This paper compares several strategies for air-fuel ratio tran-sient control. The strategies are: A factory-standard look-up table based system (a SAAB Trionic 5), a feedback PI controller with and without feed-forward throttle correction, a linear feed-forward control algorithm, and two nonlinear feed- forward algorithms based on artificial neural networks. The control strategies have been implemented and evaluated in a SAAB 9000 car during a transient driving test, consisting of an acceleration in the second gear from an engine speed of 1500 rpm to 3000 rpm. The best strategies are found to be the neural network based ones, followed by the table based factory system. The two feedback PI controllers offer the poorest performance.

This paper introduces an unsupervised method for the classification of discrete rovers' slip events based on proprioceptive signals. In particular, the method is able to automatically discover and track various degrees of slip (i.e. low slip, moderate slip, high slip). The proposed method is based on aggregating the data over time, since high level concepts, such as high and low slip, are concepts that are dependent on longer time perspectives. Different features and subsets of the data have been identified leading to a proper clustering, interpreting those clusters as initial models of the prospective concepts. Bayesian tracking has been used in order to continuously improve the parameters of these models, based on the new data. Two real datasets are used to validate the proposed approach in comparison to other known unsupervised and supervised machine learning methods. The first dataset is collected by a single-wheel testbed available at MIT. The second dataset was collected by means of a planetary exploration rover in real off-road conditions. Experiments prove that the proposed method is more accurate (up to 86% of accuracy vs. 80% for K-means) in discovering various levels of slip while being fully unsupervised (no need for hand-labeled data for training). © 2017 ISTVS

Reducing emissions and improving fuel efficiency in automobiles are today important issues. New sensor techniques are developed to extract detailed combustion information to enable closed loop engine control. This thesis is about a virtual sensor; measuring an ion current inside the cylinder by using the already existing spark plug, followed by signal processing for estimation of combustion parameters. First, the thesis aims to show that the ion current signal can be used for closed loop control of Exhaust Gas Recirculation (EGR). Use of EGR is very common in modern automobiles because of the potential reduction of NOx emissions and fuel consumption, but using too much EGR can have the reverse effect (e.g. increased fuel consumption and driveability problems). Algorithms for estimating combustion variability are proposed and a closed loop scheme for controlling an EGR valve is demonstrated for driving on the highway in a SAAB 9000. Estimation of the pressure peak position is treated for closed loop control of ignition timing. Such estimation can be performed with the ion current but may not work if a fuel additive is used. Different methods are compared and it is shown that using a fuel additive may even improve the estimation accuracy of the pressure peak position with about 25%. An algorithm is also proposed to estimate the pressure peak position even in presence of EGR. Strategies for transient control of the air-fuel ratio are also compared. Air-fuel ratio control is important because even small deviations from the stoichiometric value can result in significantly increased emissions. It is found that a neural network based controller had the best performance with approximately 23% lower RMS error than the adapted standard control module.

Two virtual sensors are proposed that use the spark-plug based ion current sensor for combustion engine control. The first sensor estimates combustion variability for the purpose of controlling exhaust gas recirculation (EGR) and the second sensor estimates the pressure peak position for control of ignition timing. Use of EGR in engines is important because the technique can reduce fuel consumption and NOx emissions, but recirculating too much can have the adverse effect with e.g. increased fuel consumption and poor driveability of the vehicle. Since EGR also affects the phasing of the combustion (because of the diluted gas mixture with slower combustion) it is also necessary to control ignition timing otherwise efficiency will be lost. The combustion variability sensor is demonstrated in a closed-loop control experiment of EGR on the highway and the pressure peak sensor is shown to handle both normal and an EGR condition.

It is desirable for an engine control system to maintain a stable combustion. A high combustion variability (typically measured by the relative variations in produced work, COV(IMEP)) can indicate the use of too much EGR or a too lean air-fuel mixture, which results in less engine efficiency(in terms of fuel and emissions) and reduced driveability. The coefficient of variation (COV) of the ion current integral has previously been shown in several papers to be correlated to the coefficient of variation of IMEP for various disturbances (e.g. AFR, EGR and fuel timing). This paper presents a cycle-to-cycle ion current based method of estimating the approximate category of IMEP (either normal burn, slow burn, partial burn or misfire) for the case of lean air-fuel ratio. The rate of appearance of the partial burn and misfire categories is then shown to be well correlated with the onset of high combustion variability(high COV(IMEP)). It is demonstrated that the detection of these categories can result in faster determination(prediction) of high variability compared to only using the COV(Ion integral). Copyright © 2004 SAE International.

Estimation of combustion variability can be performed by using ion currents measured at the spark plug. A scheme is here proposed that exploits the potential of using measurements from multiple cylinders to improve the estimation accuracy of combustion variability (measured by the coefficient of variation of IMEP). This is realised by dividing combustion variability into categories and having one classifier running for each cylinder with the ion current as input signal. The final estimate of combustion variability is then formed by a majority vote among the classifiers. This scheme is shown to improve estimation accuracy by up to 15% on measurements taken from highway driving in a production vehicle.

Fleets of commercial vehicles represent an excellent real life setting for ubiquitous knowledge discovery. There are many electronic control units onboard a modern bus or truck, with hundreds of signals being transmitted between them on the controller area network. The growing complexity of the vehicles has lead to a significant desire to have systems for fault detection, remote diagnostics and maintenance prediction. This paper aims to show that it is possible to discover useful diagnostic knowledge by a self-organized algorithm in the scenario of a fleet of city buses. The approach is demonstrated as a process consisting of two parts; Unsupervised modeling (where interesting features are discovered) and Guided search (where the previously found features are coupled to additional information sources). The modeling part searches for simple linear models in a group of vehicles, where interesting features are selected based on both non-randomness in relations and variability in the group. It is shown in an eight months long data collection study that this approach was able to discover features related to broken wheelspeed sensors. Strikingly, deviations in these features (for the vehicles with broken sensors) can be observed up to several months before a breakdown occur. This potentially allows for sufficient time to schedule the vehicle for maintenance and prepare the workshop with relevant components. © 2013 IEEE.

Methods for equipment monitoring are traditionally constructed from specific sensors and/or knowledge collected prior to implementation on the equipment. A different approach is presented here that builds up knowledge over time by exploratory search among the signals available on the internal field-bus system and comparing the observed signal relationships among a group of equipment that perform similar tasks. The approach is developed for the purpose of increasing vehicle uptime, and is therefore demonstrated in the case of a city bus and a heavy duty truck. However, it also works fine for smaller mechatronic systems like computer hard-drives. The approach builds on an onboard self-organized search for models that capture relations among signal values on the vehicles’ data buses, combined with a limited bandwidth telematics gateway and an off-line server application where the parameters of the self-organized models are compared. The presented approach represents a new look at error detection in commercial mechatronic systems, where the normal behavior of a system is actually found under real operating conditions, rather than the behavior observed in a number of laboratory tests or test-drives prior to production of the system. The approach has potential to be the basis for a self-discovering system for general purpose fault detection and diagnostics.

With the introduction of low-cost wireless communication many new applications have been made possible; applications where systems can collaboratively learn and get wiser without human supervision. One potential application is automated monitoring for fault isolation in mobile mechatronic systems such as commercial vehicles. The paper proposes an agent design that is based on uploading software agents to a fleet of mechatronic systems. Each agent searches for interesting state representations of a system and reports them to a central server application. The states from the fleet of systems can then be used to form a consensus from which it can be possible to detect deviations and even locating a fault.

Quality and up-time management of vehicles is today receiving much attention from vehicle manufacturers. One of the reasons is that there is a desire to avoiding on-road failures to addressing potential issues during routine maintenance intervals or at times more convenient to the operator. Forthcoming telematic platforms and advanced diagnostic algorithms can enable the possibility to proactively handle problems and minimize stops. The platforms bring the possibility of increasing knowledge of fault characteristics and making diagnostic decisions by using a population of vehicles. However, this requires real-time diagnostic algorithms that process data both onboard and offboard at a central server. The paper presents a self organizing approach for failure and deviation detection on a fleet of vehicles. The approach builds on using parametric models for encoding the characteristical relations between different sensor readings for a vehicle sub-system or component. The models are low-dimensional representations of the operating characteristics of a sub-system or component and are possible to transfer over a limited wireless communication channel. The approach is demonstrated on simulated data of an electronically controlled suspension system for detecting a slow valve and a leaking bellow.

Operators of fleets of vehicles desire the best possible availability and usage of their vehicles. This means the preference is that maintenance of a vehicle is scheduled with as long intervals as possible. However, it is then important to be able to detect if a component in a specific vehicle is not functioning properly earlier than expected (due to e.g. manufacturing variations). This paper proposes a telematic based fault detection scheme for enabling fault detection for diagnostics by using a population of vehicles. The basic idea is that it is possible to create low-dimensional representations of a sub-system or component in a vehicle, where the representation (or model parameters) of a vehicle can be monitored for changes compared to the model parameters observed in a fleet of vehicles. If a model in a vehicle is found to deviate compared to a group of models from a fleet of vehicles, then the vehicle is judged to need diagnostics for that component (assuming the deviation in the model cannot be attributed to e.g. a different driver behavior). The representation should be low-dimensional so it is possible to have it transferred over a limited wireless communication channel to a communications center where the comparison is made. The algorithm is shown to be able to detect leakage on simulated data from a cooling system, work is currently in progress for detecting other types of faults.

Creating fault detection software for complex mechatronic systems (e.g. modern vehicles) is costly both in terms of engineer time and hardware resources. With the availability of wireless communication in vehicles, information can be transmitted from vehicles to allow historical or fleet comparisons. New networked applications can be created that, e.g., monitor if the behavior of a certain system in a vehicle deviates compared to the system behavior observed in a fleet. This allows a new approach to fault detection that can help reduce development costs of fault detection software and create vehicle individual service planning. The COSMO (consensus self-organized modeling) methodology described in this paper creates a compact representation of the data observed for a subsystem or component in a vehicle. A representation that can be sent to a server in a backoffice and compared to similar representations for other vehicles. The backoffice server can collect representations from a single vehicle over time or from a fleet of vehicles to define a norm of the vehicle condition. The vehicle condition can then be monitored, looking for deviations from the norm. The method is demonstrated for measurements made on a real truck driven in varied conditions with ten different generated faults. The proposed method is able to detect all cases without prior information on what a fault looks like or which signals to use.

This paper investigates the use of the ionization current to estimate the Coefficient of Variation for the Indicated Mean Effective Pressure, COV(IMEP), which is a common variable for combustion stability in a spark-ignited engine. Stable combustion in this definition implies that the variance of the produced work, measured over a number of consecutive combustion cycles, is small compared to the mean of the produced work. The COV(IMEP) is varied experimentally either by increasing EGR flow or by changing the air-fuel ratio, in both a laboratory setting (engine in dynamometer) and in an on-road setting. The experiments show a positive correlation between COV(Ion integral), the Coefficient of Variation for the integrated Ion Current, and COV(IMEP), when measured under low load on an engine in a dynamometer, but not under high load conditions. On-road experiments show a positive correlation, but only in the EGR and the lean burn case. An approach based on individual cycle classification for real-time estimation of combustion stability is discussed. © Copyright 2001 Society of Automotive Engineers, Inc.

With the introduction of unleaded gasoline, special fuel agents have appeared on the market for lubricating and cleaning the valve seats. These fuel agents often contain alkali metals that have a significant impact on the ion current signal, thus affecting strategies that use the ion current for engine control and diagnosis, e.g., for estimating the location of the pressure peak. This paper introduces a method for making neural network algorithms robust to expected disturbances in the input signal and demonstrates how well this method applies to the case of disturbances to the ion current signal due to fuel additives containing sodium. The performance of the neural estimators is compared to a Gaussian fit algorithm, which they outperform. It is also shown that using a fuel additive significantly improves the estimation of the location of the pressure peak. © 2001 Society of Automotive Engineers, Inc.

Performance evaluation and anomaly detection in complex systems are time consuming tasks based on analyzing, similarity analysis and classification of many different data sets from real operations. This paper presents an original computational technology for unsupervised incremental classification of large data sets by using a specially introduced similarity analysis method. First of all the so called compressed data models are obtained from the original large data sets by a newly proposed sequential clustering algorithm. Then the datasets are compared by pairs not directly, but by using their respective compressed data models. The evaluation of the pairs is done by a special similarity analysis method that uses the so called Intelligent Sensors (Agents) and data potentials. Finally a classification decision is generated by using a predefined threshold of similarity. The applicability of the proposed computational scheme for anomaly detection, based on many available large data sets is demonstrated on an example of 18 synthetic data sets. Suggestions for further improvements of the whole computation technology and a better applicability are also discussed in the paper.

The relation between heat demand and outdoor temperature (heat power signature) is a typical feature used to diagnose abnormal heat demand. Prior work is mainly based on setting thresholds, either statistically or manually, in order to identify outliers in the power signature. However, setting the correct threshold is a difficult task since heat demand is unique for each building. Too loose thresholds may allow outliers to go unspotted, while too tight thresholds can cause too many false alarms.

Moreover, just the number of outliers does not reflect the dispersion level in the power signature. However, high dispersion is often caused by fault or configuration problems and should be considered while modeling abnormal heat demand.

In this work, we present a novel method for ranking substations by measuring both dispersion and outliers in the power signature. We use robust regression to estimate a linear regression model. Observations that fall outside of the threshold in this model are considered outliers. Dispersion is measured using coefficient of determination R2 which is a statistical measure of how close the data are to the fitted regression line.

Our method first produces two different lists by ranking substations using number of outliers and dispersion separately. Then, we merge the two lists into one using the Borda Count method. Substations appearing on the top of the list should indicate higher abnormality in heat demand compared to the ones on the bottom. We have applied our model on data from substations connected to two district heating networks in the south of Sweden. Three different approaches i.e. outlier-based, dispersion-based and aggregated methods are compared against the rankings based on return temperatures. The results show that our method significantly outperforms the state-of-the-art outlier-based method. © 2018 The Authors. Published by Elsevier Ltd.

Convolutional Neural Networks (CNNs) have shown their effectiveness in a variety of imaging applications including medical imaging diagnostics. However, these deep learning models are data-hungry and need enough labeled samples for the training phase which is limited in the medical domain. Transfer learning is one possible solution to this challenge with training a new model. Assessing model performance should be done not only based on criteria like accuracy, and area under the ROC curve, but also it is important to investigate what regions were of most interest for the classification decisions, especially for medical applications. We performed a case study on neurodegenerative disorders, in specific Alzheimer’s disease, mild cognitive im- pairment, dementia with lewy bodies and cognitively normal brains using 3D 18F-FDG-PET brain scans. Two transfer learning models, InceptionV3 and ResNet50, as well as a custom 3D-CNN that is trained from scratch are compared. Two XAI methods, occlusion and Grad-CAM are chosen to visualize the important brain regions using correctly classified cases. We found that the TL models learn significantly different decision surfaces than the 3D-CNN model. The 3D spatial structure of the brain regions are better kept in the 3D-CNN model, and that might explain the higher performance of this model over 2D-TL models. Moreover, we found out the two XAI methods provide different results, where occlusion method focused more on specific brain regions.

Purpose: The purpose of this study is to develop and validate a 3D deep learning model that predicts the final clinical diagnosis of Alzheimer's disease (AD), dementia with Lewy bodies (DLB), mild cognitive impairment due to Alzheimer's disease (MCI-AD), and cognitively normal (CN) using fluorine 18 fluorodeoxyglucose PET (18F-FDG PET) and compare model's performance to that of multiple expert nuclear medicine physicians' readers.

Materials and methods: Retrospective 18F-FDG PET scans for AD, MCI-AD, and CN were collected from Alzheimer's disease neuroimaging initiative (556 patients from 2005 to 2020), and CN and DLB cases were from European DLB Consortium (201 patients from 2005 to 2018). The introduced 3D convolutional neural network was trained using 90% of the data and externally tested using 10% as well as comparison to human readers on the same independent test set. The model's performance was analyzed with sensitivity, specificity, precision, F1 score, receiver operating characteristic (ROC). The regional metabolic changes driving classification were visualized using uniform manifold approximation and projection (UMAP) and network attention.

Results: The proposed model achieved area under the ROC curve of 96.2% (95% confidence interval: 90.6-100) on predicting the final diagnosis of DLB in the independent test set, 96.4% (92.7-100) in AD, 71.4% (51.6-91.2) in MCI-AD, and 94.7% (90-99.5) in CN, which in ROC space outperformed human readers performance. The network attention depicted the posterior cingulate cortex is important for each neurodegenerative disease, and the UMAP visualization of the extracted features by the proposed model demonstrates the reality of development of the given disorders.

Conclusion: Using only 18F-FDG PET of the brain, a 3D deep learning model could predict the final diagnosis of the most common neurodegenerative disorders which achieved a competitive performance compared to the human readers as well as their consensus. © 2021. The Author(s).

The ever increasing complexity of modern systems and equipment make the task of monitoring their health quite challenging. Traditional methods such as expert defined thresholds, physics based models and process history based techniques have certain drawbacks. Thresholds defined by experts require deep knowledge about the system and are often too conservative. Physics driven approaches are costly to develop and maintain. Finally, process history based models require large amount of data that may not be available at design time of a system. Moreover, the focus of these traditional approaches has been system specific. Hence, when industrial systems are deployed on a large scale, their monitoring becomes a new challenge. Under these conditions, this paper demonstrates the use of a group-based selfmonitoring approach that learns over time from similar systems subject to similar conditions. The approach is based on conformal anomaly detection coupled with an exchangeability test that uses martingales. This allows setting a threshold value based on sound theoretical justification. A hypothesis test based on this threshold is used to decide on if a system has deviated from its group. We demonstrate the feasibility of this approach through a real case study of monitoring a group of heat-pumps where it can detect a faulty hot-water switch-valve and a broken outdoor temperature sensor without previously observing these faults.

The monitoring infrastructure of an industrial fleet can rely on the so-called unit-level and subfleet-level models to observe the behavior of a target unit. However, such infrastructure has to confront several challenges. First, from an anomaly detection perspective of monitoring a target unit, unit-level and subfleet-level models can give different information about the nature of an anomaly, and which approach or level model is appropriate is not always clear. Second, in the absence of well-understood prior models of unit and subfleet behavior, the choice of a base model at their respective levels, especially in an online/streaming setting, may not be clear. Third, managing false alarms is a major problem. To deal with these challenges, we proposed to rely on the conformal anomaly detection framework. In addition, an ensemble approach was deployed to mitigate the knowledge gap in understanding the underlying data-generating process at the unit and subfleet levels. Therefore, to monitor the behavior of a target unit, a unit-level ensemble model (ULEM) and a subfleet-level ensemble model (SLEM) were constructed, where each member of the respective ensemble is based on a conformal anomaly detector (CAD). However, since the information obtained by these two ensemble models through their p-values may not always agree, a combined ensemble model (CEM) was proposed. The results are based on real-world operational data obtained from district heating (DH) substations. Here, it was observed that CEM reduces the overall false alarms compared to ULEM or SLEM, albeit at the cost of some detection delay. The analysis demonstrated the advantages and limitations of ULEM, SLEM, and CEM. Furthermore, discords obtained from the state-of-the-art matrix-profile (MP) method and the combined calibration scores obtained from ULEM and SLEM were compared in an offline setting. Here, it was observed that SLEM achieved a better overall precision and detection delay. Finally, the different components related to ULEM, SLEM, and CEM were put together into what we refer to as TRANTOR: a conformal anomaly detection based industrial fleet monitoring framework. The proposed framework is expected to enable fleet operators in various domains to improve their monitoring infrastructure by efficiently detecting anomalous behavior and controlling false alarms at the target units. © 2022

We considered the case of monitoring a large fleet where heterogeneity in the operational behavior among its constituent units (i.e., systems or machines) is non-negligible, and no labeled data is available. Each unit in the fleet, referred to as a target, is tracked by its sub-fleet. A conformal sub-fleet (CSF) is a set of units that act as a proxy for the normal operational behavior of a target unit by relying on the Mondrian conformal anomaly detection framework. Two approaches, the k-nearest neighbors and conformal clustering, were investigated for constructing such a sub-fleet by formulating a stability criterion. Moreover, it is important to discover the sub-sequence of events that describes an anomalous behavior in a target unit. Hence, we proposed to extract such sub-sequences for further investigation without pre-specifying their length. We refer to it as a conformal anomaly sequence (CAS). Furthermore, different nonconformity measures were evaluated for their efficiency, i.e., their ability to detect anomalous behavior in a target unit, based on the length of the observed CAS and the S-criterion value. The CSF approach was evaluated in the context of monitoring district heating substations. Anomalous behavior sub-sequences were corroborated with the domain expert leading to the conclusion that the proposed approach has the potential to be useful for both diagnostic and knowledge extraction purposes, especially in domains where labeled data is not available or hard to obtain. © 2021

A typical district heating (DH) network consists of hundreds, sometimes thousands, of substations. In the absence of a well-understood prior model or data labels about each substation, the overall monitoring of such large number of substations can be challenging. To overcome the challenge, an approach based on the collective operational monitoring of each substation by a local group (i.e., the reference-group) of other similar substations in the network was formulated. Herein, if a substation of interest (i.e., the target) starts to behave differently in comparison to those in its reference-group, then it was designated as an outlier. The approach was demonstrated on the monitoring of the return temperature variable for atypical and faulty operational behavior in 778 substations associated with multi-dwelling buildings. The choice of an appropriate similarity measure along with its size k were the two important factors that enables a reference-group to effectively detect an outlier target. Thus, different similarity measures and size k for the construction of the reference-groups were investigated, which led to the selection of the Euclidean distance with k = 80. This setup resulted in the detection of 77 target substations that were outliers, i.e., the behavior of their return temperature changed in comparison to the majority of those in their respective reference-groups. Of these, 44 were detected due to the local construction of the reference-groups. In addition, six frequent patterns of deviating behavior in the return temperature of the substations were identified using the reference-group based approach, which were then further corroborated by the feedback from a DH domain expert. © 2020 Elsevier Ltd

The behavior of a district heating (DH) substation has a social and operational context. The social context comes from its general usage pattern and personal requirements of building inhabitants. The operational context comes from its configuration settings which considers both the weather conditions and social requirements. The parameter estimating thermal energy demand response with respect to change in outdoor temperature conditions along with the strength of the relationship between these variables are two important measures of operational efficiency of a substation. In practice, they can be estimated using a regression model where the slope parameter measures the average response and R2 measures the strength of the relationship. These measures are also important from a monitoring perspective. However, factors related to the social context of a building and the presence of unexplained outliers can make the estimation of these measures a challenging task. Social context of a data point in DH, in many cases appears as an outlier. Data efficiency is also required if these measures are to be estimated in a timely manner. Under these circumstances, methods that can isolate and reduce the effect of outliers in a principled and data efficient manner are required. We therefore propose to use Huber regression, a robust method based on M-estimator type loss function. This method can not only identify possible outliers present in the data of each substation but also reduce their effect on the estimated slope parameter. Moreover, substations that are comparable according to certain criteria, for instance, those with almost identical energy demand levels, should have relatively similar slopes. This provides an opportunity to observe deviating substations under the assumption that comparable substations should show homogeneity in their behavior. Furthermore, the slope parameter can be compared across time to observe if the dynamics of a substation has changed. Our analysis shows that Huber regression in combination with ordinary least squares can provide reliable estimates on the operational efficiency of DH substations. © 2018 The Authors. Published by Elsevier Ltd.

This paper analyzes the advantages and limitations of known machine learning approaches to cope with the problem of incipient rover embedding detection based on propioceptive signals. In particular, two supervised learning approaches (Support Vector Machines and Feed-forward Neural Networks) are compared to two unsupervised learning approaches (K-means and Self-Organizing Maps) in order to identify various degrees of slip (e.g. low slip, moderate slip, high slip). A real dataset collected by a single-wheel testbed available at MIT has been used to validate each strategy. The SVM algorithm achieves the best performance (accuracy >95 %). However, the SOM algorithm represents a better solution in terms of accuracy and the need of hand-labeled data for training the classifier (accuracy >84 %).

A diagnosis and maintenance method, a diagnosis and maintenance assembly comprising a central server and a system, and a computer program for diagnosis and maintenance for a plurality of systems, particularly for a plurality of vehicles, wherein each system provides at least one system-related signal which provides the basis for the diagnosis and/or maintenance of/for the system are provided. The basis for diagnosis and/or maintenance is determined by determining for each system at least one relation between the system-related signals, comparing the compatible determined relations, determining for the plurality of systems based on the result of the comparison which relations are significant relations and providing a diagnosis and/or maintenance decision based on the determined significant relations.

In the era of big data, it is imperative to assist the human analyst in the endeavor to find solutions to ill-defined problems, i.e. to “detect the expected and discover the unexpected” (Yi et al., 2008). To their aid, a plethora of analysis support systems is available to the analysts. However, these support systems often lack visual and interactive features, leaving the analysts with no opportunity to guide, influence and even understand the automatic reasoning performed and the data used. Yet, to be able to appropriately support the analysts in their sense-making process, we must look at this process more closely. In this paper, we present the results from interviews performed together with data analysts from the automotive industry where we have investigated how they handle the data, analyze it and make decisions based on the data, outlining directions for the development of analytical support systems within the area. © Springer International Publishing Switzerland 2016.

Fuel consumption and vehicle breakdown depend upon the driving style of the driver, for example, hard driving style leads to more wear and consequently more failures of vehicle components. Because of this, it is important to identify and classify the driver’s driving style in order to give the driver feedback through a driver assistance system. The driver would then be able to detect and learn to avoid a driving style that is not appropriate. The input data is provided by different sensors installed in the vehicle, where different drivers and driving routes have been measured. The data is subjectively classified into two different driving styles: normal and hard. Hard driving style can be characterized, for example, by rapid acceleration and braking. Since it is not trivial to build a model which is able to distinguish hard driving from normal, a data mining approach has been employed. In the paper, several classifiers are compared (including e.g. neural networks and decision trees) and a discussion is made on the advantages and disadvantages of the different methods. Copyright © 2012 SAE International.

This paper presents a method for mining nonlinear relationships in machine data with the purpose of using such relationships to detect faults, isolate faults and predict wear and maintenance needs. The method is based on the symmetrical uncertainty measure from information theory, hierarchical clustering and self-organizing maps. It is demonstrated on synthetic data sets where it is shown to be able to detect interesting signal relations and outperform linear methods. It is also demonstrated on real data sets where it is considerably harder to select small feature sets. It is also demonstrated on the real data sets that there is information about system wear and system faults in the detected relationships. The work is part of a long-term research project with the aim to construct a self-organizing autonomic computing system for self-monitoring of mechatronic systems.

Predictive maintenance is becoming more and more important in many industries, especially taking into account the increasing focus on offering uptime guarantees to the customers. However, in automotive industry, there is a limitation on the engineering effort and sensor capabilities available for that purpose. Luckily, it has recently become feasible to analyse large amounts of data on-board vehicles in a timely manner. This allows approaches based on data mining and pattern recognition techniques to augment existing, hand crafted algorithms.

Automated deviation detection offers both broader applicability, by virtue of detecting unexpected faults and cross-analysing data from different subsystems, as well as higher sensitivity, due to its ability to take into account specifics of a selected, small set of vehicles used in a particular way under similar conditions.

In a project called Redi2Service we work towards developing methods for autonomous and unsupervised relationship discovery, algorithms for detecting deviations within those relationships (both considering different moments in time, and different vehicles in a fleet), as well as ways to correlate those deviations to known and unknown faults. In this paper we present the type of data we are working with, justify why we believe relationships between signals are a good knowledge representation, and show results of early experiments where supervised learning was used to evaluate discovered relations.

Predictive maintenance is becoming more and more important for the commercial vehicle manufactures, as focus shifts from product- to service-based operation. The idea is to provide a dynamic maintenance schedule, fulfilling specific needs of individual vehicles. Luckily, the same shift of focus, as well as technological advancements in the telecommunication area, make long-term data collection more widespread, delivering the necessary data.

We have found, however, that the standard attribute-value knowledge representation is not rich enough to capture important dependencies in this domain. Therefore, we are proposing a new rule induction algorithm, inspired by Michalski's classical AQ approach. Our method is aware that data concerning each vehicle consists of time-ordered sequences of readouts. When evaluating candidate rules, it takes into account the composite performance for each truck, instead of considering individual readouts in separation. This allows us more exibility, in particular in defining desired prediction horizon in a fuzzy, instead of crisp, manner. © 2013 The authors and IOS Press. All rights reserved.

Objective: This study aims to comprehensively review the use of graph neural networks (GNNs) for clinical risk prediction based on electronic health records (EHRs). The primary goal is to provide an overview of the state-of-the-art of this subject, highlighting ongoing research efforts and identifying existing challenges in developing effective GNNs for improved prediction of clinical risks. Methods: A search was conducted in the Scopus, PubMed, ACM Digital Library, and Embase databases to identify relevant English-language papers that used GNNs for clinical risk prediction based on EHR data. The study includes original research papers published between January 2009 and May 2023. Results: Following the initial screening process, 50 articles were included in the data collection. A significant increase in publications from 2020 was observed, with most selected papers focusing on diagnosis prediction (n = 36). The study revealed that the graph attention network (GAT) (n = 19) was the most prevalent architecture, and MIMIC-III (n = 23) was the most common data resource. Conclusion: GNNs are relevant tools for predicting clinical risk by accounting for the relational aspects among medical events and entities and managing large volumes of EHR data. Future studies in this area may address challenges such as EHR data heterogeneity, multimodality, and model interpretability, aiming to develop more holistic GNN models that can produce more accurate predictions, be effectively implemented in clinical settings, and ultimately improve patient care. © 2024 The Authors

It is difficult to implement predictive maintenance in the automotive industry as it looks today, since the sensor capabilities and engineering effort available for diagnostic purposes is limited. It is, in practice, impossible to develop diagnostic algorithms capable of detecting many different kinds of faults that would be applicable to a wide range of vehicle configurations and usage patterns. However, it is now becoming feasible to obtain and analyse on-board data on vehicles as they are being used. It makes automatic data-mining methods an attractive alternative, since they are capable of adapting themselves to specific vehicle configurations and usage. In order to be useful, though, such methods need to be able to detect interesting relations between a large number of available signals. This paper presents an unsupervised method for discovering useful relations between measured signals in a Volvo truck, both during normal operations and when a fault has occurred. The interesting relationships are found in a two-step procedure. In the first step, we identify a set of “good” models, by establishing an MSE threshold over the complete data set. In the second step, we estimate model parameters over time, in order to capture the dynamic behaviour of the system. We use two different approaches here, the LASSO method and the Recursive Least Squares filter. The usefulness of obtained relations is then evaluated using supervised learning to separate different classes of faults.

In multiple industries, including automotive one, predictive maintenance is becoming more and more important, especially since the focus shifts from product to service-based operation. It requires, among other, being able to provide customers with uptime guarantees. It is natural to investigate the use of data mining techniques, especially since the same shift of focus, as well as technological advancements in the telecommunication solutions, makes long-term data collection more widespread.

In this paper we describe our experiences in predicting compressor faults using data that is logged on-board Volvo trucks. We discuss unique challenges that are posed by the specifics of the automotive domain. We show that predictive maintenance is possible and can result in significant cost savings, despite the relatively low amount of data available. We also discuss some of the problems we have encountered by employing out-of-the-box machine learning solutions, and identify areas where our task diverges from common assumptions underlying the majority of data mining research.

Methods and results are presented for applying supervised machine learning techniques to the task of predicting the need for repairs of air compressors in commercial trucks and buses. Prediction models are derived from logged on-board data that are downloaded during workshop visits and have been collected over three years on large number of vehicles. A number of issues are identified with the data sources, many of which originate from the fact that the data sources were not designed for data mining. Nevertheless, exploiting this available data is very important for the automotive industry as means to quickly introduce predictive maintenance solutions. It is shown on a large data set from heavy duty trucks in normal operation how this can be done and generate a profit.

Random forest is used as the classifier algorithm, together with two methods for feature selection whose results are compared to a human expert. The machine learning based features outperform the human expert features, which supports the idea to use data mining to improve maintenance operations in this domain. © 2015 Elsevier Ltd.

An approach is presented and experimentally demonstrated where consensus among distributed self-organized agents is used for intelligent monitoring of mobile cyberphysical systems (in this case vehicles). The demonstration is done on test data from a 30 month long field test with a city bus fleet under real operating conditions. The self-organized models operate on-board the systems, like embedded agents, communicate their states over a wireless communication link, and their states are compared off-line to find systems that deviate from the consensus. In this way is the group (the fleet) of systems used to detect errors that actually occur. This can be used to build up a knowledge base that can be accumulated over the life-time of the systems.

Deviation detection is important for self-monitoring systems. To perform deviation detection well requires methods that, given only "normal" data from a distribution of unknown parametric form, can produce a reliable statistic for rejecting the null hypothesis, i.e. evidence for devating data. One measure of the strength of this evidence based on the data is the p-value, but few deviation detection methods utilize p-value estimation. We compare three methods that can be used to produce p-values: one class support vector machine (OCSVM), conformal anomaly detection (CAD), and a simple "most central pattern" (MCP) algorithm. The SVM and the CAD method should be able to handle a distribution of any shape. The methods are evaluated on synthetic data sets to test and illustrate their strengths and weaknesses, and on data from a real life self-monitoring scenario with a city bus fleet in normal traffic. The OCSVM has a Gaussian kernel for the synthetic data and a Hellinger kernel for the empirical data. The MCP method uses the Mahalanobis metric for the synthetic data and the Hellinger metric for the empirical data. The CAD uses the same metrics as the MCP method and has a k-nearest neighbour (kNN) non-conformity measure for both sets. The conclusion is that all three methods give reasonable, and quite similar, results on the real life data set but that they have clear strengths and weaknesses on the synthetic data sets. The MCP algorithm is quick and accurate when the "normal" data distribution is unimodal and symmetric (with the chosen metric) but not otherwise. The OCSVM is a bit cumbersome to use to create (quantized) p-values but is accurate and reliable when the data distribution is multimodal and asymmetric. The CAD is also accurate for multimodal and asymmetric distributions. The experiment on the vehicle data illustrate how algorithms like these can be used in a self-monitoring system that uses a fleet of vehicles to conduct deviation detection without supervisi- n and without prior knowledge about what is being monitored. © 2014 IEEE.

An approach for intelligent monitoring of mobile cyberphysical systems is described, based on consensus among distributed self-organised agents. Its usefulness is experimentally demonstrated over a long-time case study in an example domain: a fleet of city buses. The proposed solution combines several techniques, allowing for life-long learning under computational and communication constraints. The presented work is a step towards autonomous knowledge discovery in a domain where data volumes are increasing, the complexity of systems is growing, and dedicating human experts to build fault detection and diagnostic models for all possible faults is not economically viable. The embedded, self-organised agents operate on-board the cyberphysical systems, modelling their states and communicating them wirelessly to a back-office application. Those models are subsequently compared against each other to find systems which deviate from the consensus. In this way the group (e.g. a fleet of vehicles) is used to provide a standard, or to describe normal behaviour, together with its expected variability under particular operating conditions. The intention is to detect faults without the need for human experts to anticipate them beforehand. This can be used to build up a knowledge base that accumulates over the life-time of the systems. The approach is demonstrated using data collected during regular operation of a city bus fleet over the period of almost four years. © 2017 The Author(s)

An approach is proposed for automatic fault detection in a population of mechatronic systems. The idea is to employ self-organizing algorithms that produce low-dimensional representations of sensor and actuator values on the vehicles, and compare these low-dimensional representations among the systems. If a representation in one vehicle is found to deviate from, or to be not so similar to, the representations for the majority of the vehicles, then the vehicle is labeled for diagnostics. The presented approach makes use of principal component coding and a measure of distance between linear sub-spaces. The method is successfully demonstrated using simulated data for a commercial vehiclepsilas engine coolant system, and using real data for computer hard drives.

In recent years, the concept of lifelong learning has been emphasized in relation to higher education, with a bearing idea of the possibility for the individual for a continuous, self-motivated pursuit of gaining knowledge for both personal and professional reasons, provided by higher education institutions (HEI:s). But how can this actually be done in practice? In this paper we present an ongoing project called MAISTR, which is a collaboration between Swedish HEI:s and industry with the aim of providing a number of flexible courses within the subjects of Artificial intelligence (AI), Design, and Innovation management, for professionals. Our aim is to describe how the project is setup to create new learning opportunities, including the development process and co-creation with industry, the core structure and the pedagogical design. Furthermore, we would like to discuss both challenges and opportunities that come with this kind of project, as well as reflecting on early stage outcomes. © 2023, ICST Institute for Computer Sciences, Social Informatics and Telecommunications Engineering.

Background: In recent years, neuroimaging with deep learning (DL) algorithms have made remarkable advances in the diagnosis of neurodegenerative disorders. However, applying DL in different medical domains is usually challenged by lack of labeled data. To address this challenge, transfer learning (TL) has been applied to use state-of-the-art convolution neural networks pre-trained on natural images. Yet, there are differences in characteristics between medical and natural images, also image classification and targeted medical diagnosis tasks. The purpose of this study is to investigate the performance of specialized and TL in the classification of neurodegenerative disorders using 3D volumes of 18F-FDG-PET brain scans. Results: Results show that TL models are suboptimal for classification of neurodegenerative disorders, especially when the objective is to separate more than two disorders. Additionally, specialized CNN model provides better interpretations of predicted diagnosis. Conclusions: TL can indeed lead to superior performance on binary classification in timely and data efficient manner, yet for detecting more than a single disorder, TL models do not perform well. Additionally, custom 3D model performs comparably to TL models for binary classification, and interestingly perform better for diagnosis of multiple disorders. The results confirm the superiority of the custom 3D-CNN in providing better explainable model compared to TL adopted ones. © 2022, The Author(s).

A telematic based system for enabling automatic fault detection of a population of vehicles is proposed. To avoid sending huge amounts of data over the telematics gateway, the idea is to use low-dimensional representations of sensor values in sub-systems in a vehicle. These low-dimensional representations are then compared between similar systems in a fleet. If a representation in a vehicle is found to deviate from the group of systems in the fleet, then the vehicle is labeled for diagnostics for that subsystem. The idea is demonstrated on the engine coolant system and it is shown how this self-organizing approach can detect varying levels of clogged radiator.

A new approach is proposed for fault detection. It builds on using the relationships between sensor values on vehicles to detect deviating sensor readings and trends in the system performance. However, in contrast to previous approaches based on such sensor relations, our approach uses a fleet of vehicles to define the normal conditions and relations. The relationships between the sensors are also determined automatically in a self-organized way on each vehicle, i.e. no off-line modeling is required. The proposed method is the first step in a remote diagnostics and maintenance service where error detection is done automatically, followed by a download of special purpose diagnostics software for the particular subsystem where the possible fault was detected.

A new approach to improve fault detection is proposed. The method takes benefit of using a population of systems to dynamically define a norm of how the system works. The norm is derived from self-organizing algorithms which generate a low dimensional representation of how selected feature data are correlated. The feature data is selected from the state variables and from the control signals. The self-organizing method and limited number of feature signals enable fast deviation detection and low computational footprint on each system to be analyzed. The comparison analysis between the systems is performed at a service centre, to where the low-dimensional representations of the systems are transmitted. The method is demonstrated on a simulated DC-motor and the results are promising for deviation detection.

A new approach to improve fault detection of electrical machines is proposed. The increased usage of electrical machines and the higher demands on their availability requires new approaches to fault detection. In this paper we demonstrate that it is possible to detect a certain fault on a PMSM (Permanent Magnet Synchronous Machine) by using multiple similar motors, or a single motor, to build a norm of expected behavior by monitoring signal relations. This means that the machine is monitored in an unsupervised way. Four levels of an increased temperature in the rotor magnets have been investigated. The results are based on simulations and the signals used (for relation measurements) are available in a real motor installation. The method shows promising results in detecting two of the temperature faults. © 2011 IEEE.

The heavy vehicle industry has today no requirement to provide a tire pressure monitoring system by law. This has created issues surrounding unknown tire pressure and thread depth during active service. There is also no standardization for these kind of systems which means that different manufacturers and third party solutions work after their own principles and it can be hard to know what works for a given vehicle type. The objective is to create an indirect tire monitoring system that can generalize a method that detect both incorrect tire pressure and thread depth for different type of vehicles within a fleet without the need for additional physical sensors or vehicle specific parameters. The existing sensors that are connected communicate through CAN and are interpreted by the Drivec Bridge hardware that exist in the fleet. By using supervised machine learning a classifier was created for each axle where the main focus was the front axle which had the most issues. The classifier will classify the vehicles tires condition and will be implemented in Drivecs cloud service where it will receive its data. The resulting classifier is a random forest implemented in Python. The result from the front axle with a data set consisting of 9767 samples of buses with correct tire condition and 1909 samples of buses with incorrect tire condition it has an accuracy of 90.54% (0.96%). The data sets are created from 34 unique measurements from buses between January and May 2017. This classifier has been exported and is used inside a Node.js module created for Drivecs cloud service which is the result of the whole implementation. The developed solution is called Indirect Tire Monitoring System (ITMS) and is seen as a process. This process will predict bad classes in the cloud which will lead to warnings. The warnings are defined as incidents. They contain only the information needed and the bandwidth of the incidents are also controlled so incidents are created within an acceptable range over a period of time. These incidents will be notified through the cloud for the operator to analyze for upcoming maintenance decisions. © 2017 Published under licence by IOP Publishing Ltd.

Recent research has found deep neural networks to be vulnerable, by means of prediction error, to images corrupted by small amounts of non-random noise. These images, known as adversarial examples are created by exploiting the input to output mapping of the network. For the MNIST database, we observe in this paper how well the known regularization/robustness methods improve generalization performance of deep neural networks when classifying adversarial examples and examples perturbed with random noise. We conduct a comparison of these methods with our proposed robustness method, an ensemble of models trained on adversarial examples, able to clearly reduce prediction error. Apart from robustness experiments, human classification accuracy for adversarial examples and examples perturbed with random noise is measured. Obtained human classification accuracy is compared to the accuracy of deep neural networks measured in the same experimental settings. The results indicate, human performance does not suffer from neural network adversarial noise.

The battery cells are an important part of electric and hybrid vehicles and their deterioration due to aging directly affects the life cycle and performance of the whole battery system. Therefore an early aging detection of the battery cell is an important task and its correct solution could significantly improve the whole vehicle performance. This paper presents a computational strategy for battery aging detection, based on available data chunks from real operation of the vehicle. The first step is to aggregate (reduce) the original large amount of data by much smaller number of cluster centers. This is done by a newly proposed sequential clustering algorithm that arranges the clusters in decreasing order of their volumes. The next step is the proposed fuzzy inference procedure for weighed approximation of the cluster centers that creates comparable one dimensional fuzzy model for each available data set. Finally, the detection of the aged battery is treated as a similarity analysis problem, in which the pair distances between all battery cells are estimated by analyzing the predicted values from the respective fuzzy models. All these three steps of the computational procedure are explained in the paper and applied to real experimental data for battery aging detection. The results are positive and suggestions for further improvements are made in the conclusions. © 2012 IEEE.

The battery cells are an important part of electric and hybrid vehicles, and their deterioration due to aging or malfunction directly affects the life cycle and performance of the whole battery system. Therefore, an early detection of deviation in performance of the battery cells is an important task and its correct solution could significantly improve the whole vehicle performance. This paper presents a computational strategy for the detection of deviation of battery cells, due to aging or malfunction. The detection is based on periodically processing a predetermined number of data collected in data blocks that are obtained during the real operation of the vehicle. The first step is data compression, when the original large amount of data is reduced to smaller number of cluster centers. This is done by a newly proposed sequential clustering algorithm that arranges the clusters in decreasing order of their volumes. The next step is using a fuzzy inference procedure for weighted approximation of the cluster centers to create one-dimensional models for each battery cell that represents the voltage–current relationship. This creates an equal basis for the further comparison of the battery cells. Finally, the detection of the deviated battery cells is treated as a similarity-analysis problem, in which the pair distances between all battery cells are estimated by analyzing the estimations for voltage from the respective fuzzy models. All these three steps of the computational procedure are explained in the paper and applied to real experimental data for the detection of deviation of five battery cells. Discussions and suggestions are made for a practical application aimed at designing a monitoring system for the detection of deviations. © 2013 Wiley Periodicals, Inc.