Pub Date : 2022-06-29DOI: 10.36001/phme.2022.v7i1.3361
Yuning He, J. Schumann, Huafeng Yu
AI components (e.g., Deep Neural Networks) are increasingly used in safety-relevant aerospace applications. Rigorous Verification and Validation (V&V) is mandatory for such components, yet V&V techniques for DNNs are still in their infancy and can often only provide relatively weak guarantees. In this paper, we will present a runtime-monitoring architecture, which combines the advanced statistical analysis framework SYSAI (System Analysis using Statistical AI) with temporal and probabilistic runtime monitoring carried out by R2U2 (Realizable, Responsive, and Unobtrusive Unit). We will present initial results of our tool set and architecture on a case study, a DNN-based autonomous centerline tracking system (ACT).
{"title":"Toward Runtime Assurance of Complex Systems with AI Components","authors":"Yuning He, J. Schumann, Huafeng Yu","doi":"10.36001/phme.2022.v7i1.3361","DOIUrl":"https://doi.org/10.36001/phme.2022.v7i1.3361","url":null,"abstract":"AI components (e.g., Deep Neural Networks) are increasingly used in safety-relevant aerospace applications. Rigorous Verification and Validation (V&V) is mandatory for such components, yet V&V techniques for DNNs are still in their infancy and can often only provide relatively weak guarantees. In this paper, we will present a runtime-monitoring architecture, which combines the advanced statistical analysis framework SYSAI (System Analysis using Statistical AI) with temporal and probabilistic runtime monitoring carried out by R2U2 (Realizable, Responsive, and Unobtrusive Unit). We will present initial results of our tool set and architecture on a case study, a DNN-based autonomous centerline tracking system (ACT).","PeriodicalId":422825,"journal":{"name":"PHM Society European Conference","volume":"4 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128303446","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-29DOI: 10.36001/phme.2022.v7i1.3341
P. Bischof, F. Thielecke, D. Metzler
Hydraulic systems in conventional civil aviation are currently monitored in a very rudimentary way. Normally, measured values are compared with a fixed threshold. If these measured values are outside the predefined limits, the entire hydraulic system is usually shut down. To overcome this deficit, a study regarding a novel prognostic health management method for aircraft hydraulic pumps, which allows a statement about the pump condition, is presented in this paper. The method is based on measuring differential pressure and temperature at a suitable resistance. In the first part of the study, the overall concept for monitoring the motor pump unit is analyzed. This is followed by a discussion of possible measurement methods and suitable resistors to determine the condition of the pump. In the second part of the study, the implementation for online monitoring of the pump is discussed. After a suitable approximation is found, the quality of the proposed method is evaluated with real hydraulic power generation and consumers.
{"title":"Online Flow Estimation for Condition Monitoring of Pumps in Aircraft Hydraulics","authors":"P. Bischof, F. Thielecke, D. Metzler","doi":"10.36001/phme.2022.v7i1.3341","DOIUrl":"https://doi.org/10.36001/phme.2022.v7i1.3341","url":null,"abstract":"Hydraulic systems in conventional civil aviation are currently monitored in a very rudimentary way. Normally, measured values are compared with a fixed threshold. If these measured values are outside the predefined limits, the entire hydraulic system is usually shut down. To overcome this deficit, a study regarding a novel prognostic health management method for aircraft hydraulic pumps, which allows a statement about the pump condition, is presented in this paper. The method is based on measuring differential pressure and temperature at a suitable resistance. In the first part of the study, the overall concept for monitoring the motor pump unit is analyzed. This is followed by a discussion of possible measurement methods and suitable resistors to determine the condition of the pump. In the second part of the study, the implementation for online monitoring of the pump is discussed. After a suitable approximation is found, the quality of the proposed method is evaluated with real hydraulic power generation and consumers.","PeriodicalId":422825,"journal":{"name":"PHM Society European Conference","volume":"21 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132490549","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-29DOI: 10.36001/phme.2022.v7i1.3311
Piero Danti, Ryota Minamino, G. Vichi
In the last ten years, Machine Learning (ML) and Artificial Intelligence (AI) have overwhelmed every engineering research branch finding a broad variety of applications; anomaly detection and anomaly classification are two of the topics that have benefited mostly by data-driven methods’ insights. On the other side, in the small diesel engine domain, the current trend is to lean on traditional anomaly detection/classification procedures and do not foster the use of AI. The goal of this work is to detect anomalies in the in-cylinders injectors of a small diesel engine as soon as a wrong quantity of fuel is inputted into one or more cylinders by means of ML approaches. Part of the analysis aim to understand which measurements are the most relevant for the detection and to compare different techniques to select the most suitable one. Furthermore, a condition-based methodology for maintenance is proposed. After a brief review of the state-of-the-art, the case study scenario is presented grouping sensors accordingly to their degree of accessibility; then, the implemented techniques are explained, and results are discussed.
{"title":"Wrong Injection Detection in a Small Diesel Engine, a Machine Learning Approach","authors":"Piero Danti, Ryota Minamino, G. Vichi","doi":"10.36001/phme.2022.v7i1.3311","DOIUrl":"https://doi.org/10.36001/phme.2022.v7i1.3311","url":null,"abstract":"In the last ten years, Machine Learning (ML) and Artificial Intelligence (AI) have overwhelmed every engineering research branch finding a broad variety of applications; anomaly detection and anomaly classification are two of the topics that have benefited mostly by data-driven methods’ insights. On the other side, in the small diesel engine domain, the current trend is to lean on traditional anomaly detection/classification procedures and do not foster the use of AI. The goal of this work is to detect anomalies in the in-cylinders injectors of a small diesel engine as soon as a wrong quantity of fuel is inputted into one or more cylinders by means of ML approaches. Part of the analysis aim to understand which measurements are the most relevant for the detection and to compare different techniques to select the most suitable one. Furthermore, a condition-based methodology for maintenance is proposed. After a brief review of the state-of-the-art, the case study scenario is presented grouping sensors accordingly to their degree of accessibility; then, the implemented techniques are explained, and results are discussed.","PeriodicalId":422825,"journal":{"name":"PHM Society European Conference","volume":"109 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127665450","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Quantifying the predictive uncertainty of a model is an important ingredient in data-driven decision making. Uncertainty quantification has been gaining interest especially for deep learning models, which are often hard to justify or explain. Various techniques for deep learning based uncertainty estimates have been developed primarily for image classification and segmentation, but also for regression and forecasting tasks. Uncertainty quantification for anomaly detection tasks is still rather limited for image data and has not yet been demonstrated for machine fault detection in PHM applications. In this paper we suggest an approach to derive an uncertaintyinformed anomaly score for regression models trained with normal data only. The score is derived using a deep ensemble of probabilistic neural networks for uncertainty quantification. Using an example of wind-turbine fault detection, we demonstrate the superiority of the uncertainty-informed anomaly score over the conventional score. The advantage is particularly clear in an ”out-of-distribution” scenario, in which the model is trained with limited data which does not represent all normal regimes that are observed during model deployment.
{"title":"Uncertainty Informed Anomaly Scores with Deep Learning: Robust Fault Detection with Limited Data","authors":"Jannik Zgraggen, Gianmarco Pizza, Lilach Goren Huber","doi":"10.36001/phme.2022.v7i1.3342","DOIUrl":"https://doi.org/10.36001/phme.2022.v7i1.3342","url":null,"abstract":"Quantifying the predictive uncertainty of a model is an important ingredient in data-driven decision making. Uncertainty quantification has been gaining interest especially for deep learning models, which are often hard to justify or explain. Various techniques for deep learning based uncertainty estimates have been developed primarily for image classification and segmentation, but also for regression and forecasting tasks. Uncertainty quantification for anomaly detection tasks is still rather limited for image data and has not yet been demonstrated for machine fault detection in PHM applications.\u0000In this paper we suggest an approach to derive an uncertaintyinformed anomaly score for regression models trained with normal data only. The score is derived using a deep ensemble of probabilistic neural networks for uncertainty quantification. Using an example of wind-turbine fault detection, we demonstrate the superiority of the uncertainty-informed anomaly score over the conventional score. The advantage is particularly clear in an ”out-of-distribution” scenario, in which the model is trained with limited data which does not represent all normal regimes that are observed during model deployment.","PeriodicalId":422825,"journal":{"name":"PHM Society European Conference","volume":"7 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128076197","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-29DOI: 10.36001/phme.2022.v7i1.3315
A. Fentaye, V. Zaccaria, K. Kyprianidis
Gas turbine sensors are prone to bias and drift. They may also become unavailable due to maintenance activities or failure through time. It is, therefore, important to correct faulty signal or replace missing sensors with estimated values for improved diagnostic solutions. Coping with a small number of sensors is the most difficult to achieve since this often leads to underdetermined and indistinguishable diagnostic problems in multiple fault scenarios. On the other hand, installing additional sensors has been a controversial issue from cost and weight perspectives. Gas path locations with difficult conditions to install sensors is also among other sensor installation related challenges. This paper proposes a sensor fault/failure correction and missing sensor replacement method. Auto-regressive integrated moving average models are employed to correct measurements from faulty and failed sensors. To replace additional sensors needed for further diagnostic accuracy improvements, neural network models are devised. The performance of the developed approach is demonstrated by applying to a three-shaft turbofan engine. Test results verify that the method proposed can well-recover measurements from faulty/failed sensors, no matter with small or major failures. It can also compensate key missing temperature and pressure measurements on the gas path based on the data from other available sensors.
{"title":"Sensor Fault/Failure Correction and Missing Sensor Replacement for Enhanced Real-time Gas Turbine Diagnostics","authors":"A. Fentaye, V. Zaccaria, K. Kyprianidis","doi":"10.36001/phme.2022.v7i1.3315","DOIUrl":"https://doi.org/10.36001/phme.2022.v7i1.3315","url":null,"abstract":"Gas turbine sensors are prone to bias and drift. They may also become unavailable due to maintenance activities or failure through time. It is, therefore, important to correct faulty signal or replace missing sensors with estimated values for improved diagnostic solutions. Coping with a small number of sensors is the most difficult to achieve since this often leads to underdetermined and indistinguishable diagnostic problems in multiple fault scenarios. On the other hand, installing additional sensors has been a controversial issue from cost and weight perspectives. Gas path locations with difficult conditions to install sensors is also among other sensor installation related challenges. This paper proposes a sensor fault/failure correction and missing sensor replacement method. Auto-regressive integrated moving average models are employed to correct measurements from faulty and failed sensors. To replace additional sensors needed for further diagnostic accuracy improvements, neural network models are devised. The performance of the developed approach is demonstrated by applying to a three-shaft turbofan engine. Test results verify that the method proposed can well-recover measurements from faulty/failed sensors, no matter with small or major failures. It can also compensate key missing temperature and pressure measurements on the gas path based on the data from other available sensors.","PeriodicalId":422825,"journal":{"name":"PHM Society European Conference","volume":"9 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122182781","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-29DOI: 10.36001/phme.2022.v7i1.3314
Sangho Lee, Youngdoo Son, Chihyeon Choi
The feature-based methods for bearing fault diagnosis in prognostics and health management have been achieved satisfactory performances because of their robustness to noise and reduced dimension by pre-defined features. However, widely employed time- and frequency-domain features are insufficient to recognize the global pattern that indicates the structure of a time-series instance. In this paper, we propose two novel graph-based features which reflect the connection strength and degree of time series, respectively. First, we construct a graph of which an edge is defined as the Euclidean distance between each pair of time steps to measure the strengths of connections between the nodes. The other graph is constructed by the visibility algorithm, which converts a time series into a complex network to reflect the degrees of connections. Then, we calculate the Frobenius norms of the adjacency matrices of both graphs and use them as features for bearing fault diagnosis. To verify the proposed features, we performed several experiments with both synthetic and real datasets. From the synthetic datasets, it is observed that the changes in amplitudes and frequencies are detected by the features for the connection strength and degree, respectively. In addition, the proposed features also well-separate the distributions of each bearing state, including normal and several fault types, and show significant performance improvement as applied to the fault diagnosis task.
{"title":"Novel Graph-Based Features for Bearing Fault Diagnosis: Two Aspects of Time Series Structure","authors":"Sangho Lee, Youngdoo Son, Chihyeon Choi","doi":"10.36001/phme.2022.v7i1.3314","DOIUrl":"https://doi.org/10.36001/phme.2022.v7i1.3314","url":null,"abstract":"The feature-based methods for bearing fault diagnosis in prognostics and health management have been achieved satisfactory performances because of their robustness to noise and reduced dimension by pre-defined features. However, widely employed time- and frequency-domain features are insufficient to recognize the global pattern that indicates the structure of a time-series instance. In this paper, we propose two novel graph-based features which reflect the connection strength and degree of time series, respectively. First, we construct a graph of which an edge is defined as the Euclidean distance between each pair of time steps to measure the strengths of connections between the nodes. The other graph is constructed by the visibility algorithm, which converts a time series into a complex network to reflect the degrees of connections. Then, we calculate the Frobenius norms of the adjacency matrices of both graphs and use them as features for bearing fault diagnosis. To verify the proposed features, we performed several experiments with both synthetic and real datasets. From the synthetic datasets, it is observed that the changes in amplitudes and frequencies are detected by the features for the connection strength and degree, respectively. In addition, the proposed features also well-separate the distributions of each bearing state, including normal and several fault types, and show significant performance improvement as applied to the fault diagnosis task.","PeriodicalId":422825,"journal":{"name":"PHM Society European Conference","volume":"77 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123902618","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-29DOI: 10.36001/phme.2022.v7i1.3363
C. Peeters, A. Jakobsson, J. Antoni, J. Helsen
The short-time Fourier transform (STFT) is a staple analysis tool for vibration signal processing due to it being a robust, non-parametric, and computationally efficient technique to analyze non-stationary signals. However, despite these beneficial properties, the STFT suffers from high variance, high sidelobes, and a low resolution. This paper investigates an alternative non-parametric method, namely the sliding-window iterative adaptive approach, to use for time-frequency representations of non-stationary vibrations. This method reduces the sidelobe levels and allows for high resolution estimates. The performance of the method is evaluated on both simulated and experimental vibration data of slow rotating machinery such as a multi-megawatt wind turbine gearbox. The results indicate significant benefits as compared to the STFT with regard to accuracy, readability, and versatility.
{"title":"Improved Time-Frequency Representation for Non-stationary Vibrations of Slow Rotating Machinery","authors":"C. Peeters, A. Jakobsson, J. Antoni, J. Helsen","doi":"10.36001/phme.2022.v7i1.3363","DOIUrl":"https://doi.org/10.36001/phme.2022.v7i1.3363","url":null,"abstract":"The short-time Fourier transform (STFT) is a staple analysis tool for vibration signal processing due to it being a robust, non-parametric, and computationally efficient technique to analyze non-stationary signals. However, despite these beneficial properties, the STFT suffers from high variance, high sidelobes, and a low resolution. This paper investigates an alternative non-parametric method, namely the sliding-window iterative adaptive approach, to use for time-frequency representations of non-stationary vibrations. This method reduces the sidelobe levels and allows for high resolution estimates. The performance of the method is evaluated on both simulated and experimental vibration data of slow rotating machinery such as a multi-megawatt wind turbine gearbox. The results indicate significant benefits as compared to the STFT with regard to accuracy, readability, and versatility.","PeriodicalId":422825,"journal":{"name":"PHM Society European Conference","volume":"70 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126497172","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-29DOI: 10.36001/phme.2022.v7i1.3339
Felix Mardt, P. Bischof, F. Thielecke
In a system’s design phase, where knowledge about the actual behavior of the system is shallow, the design of an efficient and robust system diagnostics is a complex task. In order to simplify this process, this paper presents a modelbased methodology for the design of fault diagnosis schemes. The methodology analyzes the structure of available behavioral models of the system and proposes minimal sets of sensors to fulfill diagnostic requirements. In order to choose an optimal set of sensors, the method evaluates the sets in terms of costs and diagnostic robustness. The proposed fault detection, isolation and identification schemes rely on the robust evaluation of model-based residuals using Monte-Carlo methods and highest density regions to account for measurement and parameter uncertainty. To show the design capabilities, the presented method is applied to an aircraft hydraulic power package and the resulting schemes are tested on a real test rig.
{"title":"Design Methodology for Robust Model-Based Fault Diagnosis Schemes and its Application to an Aircraft Hydraulic Power Package","authors":"Felix Mardt, P. Bischof, F. Thielecke","doi":"10.36001/phme.2022.v7i1.3339","DOIUrl":"https://doi.org/10.36001/phme.2022.v7i1.3339","url":null,"abstract":"In a system’s design phase, where knowledge about the actual behavior of the system is shallow, the design of an efficient and robust system diagnostics is a complex task. In order to simplify this process, this paper presents a modelbased methodology for the design of fault diagnosis schemes. The methodology analyzes the structure of available behavioral models of the system and proposes minimal sets of sensors to fulfill diagnostic requirements. In order to choose an optimal set of sensors, the method evaluates the sets in terms of costs and diagnostic robustness. The proposed fault detection, isolation and identification schemes rely on the robust evaluation of model-based residuals using Monte-Carlo methods and highest density regions to account for measurement and parameter uncertainty. To show the design capabilities, the presented method is applied to an aircraft hydraulic power package and the resulting schemes are tested on a real test rig.","PeriodicalId":422825,"journal":{"name":"PHM Society European Conference","volume":"19 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129306952","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-29DOI: 10.36001/phme.2022.v7i1.3348
Douw Marx, K. Gryllias
Early and accurate detection of rolling element bearing faults in rotating machinery is important for minimizing production downtime and reducing unnecessary preventative maintenance. Several fault detection methods based on signal processing and machine learning methods have been proposed. Particularly, supervised, data-driven approaches have proved to be very effective for fault detection and diagnostics of rolling element bearings. However, supervised methods rely heavily on the availability of failure data with volume, variety and veracity, which is mostly unavailable in industry. As an alternative data-driven strategy, unsupervised methods are trained on healthy data only and do not require any failure data. In contrast to supervised and un-supervised data-driven models, physics-based and phenomenological models are based on domain knowledge and not on historical data. Although these models are useful for studying the way in which damage is expected to manifest in a measured signal, they are difficult to calibrate and often lack the fidelity required to model reality. In this paper, an unsupervised data-driven anomaly detection method that exploits informative domain knowledge is proposed. Hereby, the versatility of unsupervised data-driven methods are combined with domain knowledge. In this approach, supplementary training data is generated by augmenting healthy data towards its possible future faulty state based on the characteristic bearing fault frequencies. Both healthy and augmented squared envelope spectrum data is used to train an autoencoder model that includes regularisation designed to constrain the latent features at the autoencoder bottleneck. Regularisation in the autoencoder loss enforces that the expected deviation of the healthy latent representation towards the augmented latent representation at dam aged conditions, is constrained to be maximally different for different fault modes. Consequently, the likelihood of a new test sample being healthy can be evaluated based on the projection of the sample onto an expected failure direction in the latent representation. A phenomenological and experimental dataset is used to demonstrate that the addition of augmented training data and a specialized autoencoder loss function can create a separable latent representation that can be used to generate interpretable health indicators.
{"title":"Domain Knowledge Informed Unsupervised Fault Detection for Rolling Element Bearings","authors":"Douw Marx, K. Gryllias","doi":"10.36001/phme.2022.v7i1.3348","DOIUrl":"https://doi.org/10.36001/phme.2022.v7i1.3348","url":null,"abstract":"Early and accurate detection of rolling element bearing faults in rotating machinery is important for minimizing production downtime and reducing unnecessary preventative maintenance. Several fault detection methods based on signal processing and machine learning methods have been proposed. Particularly, supervised, data-driven approaches have proved to be very effective for fault detection and diagnostics of rolling element bearings. However, supervised methods rely heavily on the availability of failure data with volume, variety and veracity, which is mostly unavailable in industry. As an alternative data-driven strategy, unsupervised methods are trained on healthy data only and do not require any failure data.\u0000In contrast to supervised and un-supervised data-driven models, physics-based and phenomenological models are based on domain knowledge and not on historical data. Although these models are useful for studying the way in which damage is expected to manifest in a measured signal, they are difficult to calibrate and often lack the fidelity required to model reality. In this paper, an unsupervised data-driven anomaly detection method that exploits informative domain knowledge is proposed. Hereby, the versatility of unsupervised data-driven methods are combined with domain knowledge.\u0000In this approach, supplementary training data is generated by augmenting healthy data towards its possible future faulty state based on the characteristic bearing fault frequencies. Both healthy and augmented squared envelope spectrum data is used to train an autoencoder model that includes regularisation designed to constrain the latent features at the autoencoder bottleneck. Regularisation in the autoencoder loss enforces that the expected deviation of the healthy latent representation towards the augmented latent representation at dam aged conditions, is constrained to be maximally different for different fault modes. Consequently, the likelihood of a new test sample being healthy can be evaluated based on the projection of the sample onto an expected failure direction in the latent representation.\u0000A phenomenological and experimental dataset is used to demonstrate that the addition of augmented training data and a specialized autoencoder loss function can create a separable latent representation that can be used to generate interpretable health indicators.","PeriodicalId":422825,"journal":{"name":"PHM Society European Conference","volume":"160 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116157255","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-29DOI: 10.36001/phme.2022.v7i1.3355
Kammal Al-Kahwati, W. Birk, Evert Flygel Nilsfors, R. Nilsen
Availability of belt conveyor systems is essential in production and logistic lines to safeguard production and delivery targets to customers. In this paper, experiences from commissioning, validation, and operation of an interactive predictive maintenance solution are reported. The solution and its development is formerly presented in Al-Kahwati et.al. (Al-Kahwati, Saari, Birk, & Atta, 2021), where the principles to derive a digital twin of a typical belt conveyor system comprising component-level degradation models,estimation schemes for the remaining useful life and the degradation rate, and vision-based hazardous object detection. Furthermore, the validation approach of modifying the belt conveyor and thus exploiting the idler misalignment load (IML) for the degradation predictions for individual components (including long-lasting ones) together with the actionable insights for the decision support is presented and assessed. Moreover, the approach to testing and validation of the object detection and its performance is assessed and presented in the same manner. An overall system assessment is then given and concludes the paper together with lessons learned. As pilot site for the study a belt conveyor system at LKAB Narvik in northern Norway is used.
{"title":"Experiences of a Digital Twin Based Predictive Maintenance Solution for Belt Conveyor Systems","authors":"Kammal Al-Kahwati, W. Birk, Evert Flygel Nilsfors, R. Nilsen","doi":"10.36001/phme.2022.v7i1.3355","DOIUrl":"https://doi.org/10.36001/phme.2022.v7i1.3355","url":null,"abstract":"Availability of belt conveyor systems is essential in production and logistic lines to safeguard production and delivery targets to customers. In this paper, experiences from commissioning, validation, and operation of an interactive predictive maintenance solution are reported. The solution and its development is formerly presented in Al-Kahwati et.al. (Al-Kahwati, Saari, Birk, & Atta, 2021), where the principles to derive a digital twin of a typical belt conveyor system comprising component-level degradation models,estimation schemes for the remaining useful life and the degradation rate, and vision-based hazardous object detection.\u0000Furthermore, the validation approach of modifying the belt conveyor and thus exploiting the idler misalignment load (IML) for the degradation predictions for individual components (including long-lasting ones) together with the actionable insights for the decision support is presented and assessed. Moreover, the approach to testing and validation of the object detection and its performance is assessed and presented in the same manner. An overall system assessment is then given and concludes the paper together with lessons learned.\u0000As pilot site for the study a belt conveyor system at LKAB Narvik in northern Norway is used.","PeriodicalId":422825,"journal":{"name":"PHM Society European Conference","volume":"20 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125268829","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：