Vehicle intelligence tool for early warning with fault signature
A method for early warning is provided. The method clusters normal historical data of normal cars into groups based on the car subsystem to which they belong. The method extracts (i) features based on group membership and (ii) feature correlations based on correlation graphs formed from the groups. The method trains an Auto-Encoder and Auto Decoder (AE&AD) model based on the features and the feature correlations to reconstruct the normal historical data with minimum reconstruction errors. The method reconstructs, using the trained AE&AD model, historical data of specific car fault types with reconstruction errors, normalizes the reconstruction errors, and selects features of the car faults with a top k large errors as fault signatures. The method reconstructs streaming data of monitored cars using the trained AE&AD model to determine streaming reconstruction errors, comparing the streaming reconstruction errors with the fault signatures to predict and provide alerts for impending known faults.
1. A computer-implemented method for early warning, comprising:
clustering normal historical data of normal cars into a plurality of groups based on a car subsystem to which they belong;
extracting (i) features based on membership to the plurality of groups and (ii) feature correlations based on correlation graphs formed from the plurality of groups to retrieve relevant signals and filter signal noise;
training, by a hardware processor, an Auto-Encoder and Auto Decoder (AE&AD) model based on the features and the feature correlations to reconstruct the normal historical data by minimizing reconstruction errors;
reconstructing, using the trained AE&AD model, historical data of specific car fault types with reconstruction errors, normalizing the reconstruction errors, and selecting features of the car faults with a top k large errors as fault signatures; and
reconstructing streaming data of monitored cars using the trained AE&AD model, installed in an edge module of the monitored car that is connected to an online car monitoring system, to determine streaming reconstruction errors, comparing the streaming reconstruction errors with the fault signatures to predict impending known faults and provide alerts for the impending known faults.
2. The computer-implemented method of claim 1 , further comprising switching to a backup system to bypass a malfunctioning system responsive to providing an alert for an impending fault.
3. The computer-implemented method of claim 1 , wherein the AE&AD model is trained using a recurrent neural network.
4. The computer-implemented method of claim 1 , wherein an auto-encoder portion of the AE&AD comprises a neural network for casting the features and the feature correlations into a space with reduced dimension to obtain compressed features, and wherein an auto-decoder portion of the AE&AD comprises a neural network that casts the compressed features into a space with increased dimension.
5. The computer-implemented method of claim 4 , wherein the neural network comprised in the auto-encoder portion of the AE&AD is symmetric with the neural network comprised in the auto-decoder portion of the AE&AD.
6. The computer-implemented method of claim 1 , wherein the AE&AD model comprises a Long Short-Term Memory (LSTM) for receiving the normal historical data, wherein an output of the LSTM is concatenated with a feature correlation graph as an input to an auto-encoder comprised in the AE&AD model.
7. The computer-implemented method of claim 1 , further comprising computing anomaly scores from the reconstruction errors.
8. The computer-implemented method of claim 7 , wherein the reconstruction errors are integrated by weighted sum into the anomaly scores.
9. The computer-implemented method of claim 7 , wherein the reconstruction errors that are normalized comprise a first reconstruction error for a given feature and a second reconstruction error for a feature correlations between the given feature and another feature.
10. A computer program product for early warning, the computer program product comprising a non-transitory computer readable storage medium having program instructions embodied therewith, the program instructions executable by a computer to cause the computer to perform a method comprising:
clustering, by a hardware processor of the computer, normal historical data of normal cars into a plurality of groups based on a car subsystem to which they belong;
extracting, by the hardware processor, (i) features based on membership to the plurality of groups and (ii) feature correlations based on correlation graphs formed from the plurality of groups;
training, by the hardware processor, an Auto-Encoder and Auto Decoder (AE&AD) model based on the features and the feature correlations to reconstruct the normal historical data by minimizing reconstruction errors;
reconstructing, by the hardware processor using the trained AE&AD model, historical data of specific car fault types with reconstruction errors, normalizing the reconstruction errors, and selecting features of the car faults with a top k large errors as fault signatures; and
reconstructing, by the hardware processor, streaming data of monitored cars using the trained AE&AD model, installed in an edge module of the monitored car that is connected to an online car monitoring system, to determine streaming reconstruction errors, comparing the streaming reconstruction errors with the fault signatures to predict impending known faults and provide alerts for the impending known faults.
11. The computer program product of claim 10 , further comprising switching to a backup system to bypass a malfunctioning system responsive to providing an alert for an impending fault.
12. The computer program product of claim 10 , wherein the AE&AD model is trained using a recurrent neural network.
13. The computer program product of claim 10 , wherein an auto-encoder portion of the AE&AD comprises a neural network for casting the features and the feature correlations into a space with reduced dimension to obtain compressed features, and wherein an auto-decoder portion of the AE&AD comprises a neural network that casts the compressed features into a space with increased dimension.
14. The computer program product of claim 13 , wherein the neural network comprised in the auto-encoder portion of the AE&AD is symmetric with the neural network comprised in the auto-decoder portion of the AE&AD.
15. The computer program product of claim 10 , wherein the AE&AD model comprises a Long Short-Term Memory (LSTM) for receiving the normal historical data, wherein an output of the LSTM is concatenated with a feature correlation graph as an input to an auto-encoder comprised in the AE&AD model.
16. The computer program product of claim 10 , further comprising computing anomaly scores from the reconstruction errors.
17. The computer program product of claim 16 , wherein the reconstruction errors are integrated by weighted sum into the anomaly scores.
18. The computer program product of claim 16 , wherein the reconstruction errors that are normalized comprise a first reconstruction error for a given feature and a second reconstruction error for a feature correlations between the given feature and another feature.
19. A computer processing system for early warning, comprising:
a memory device for storing program code; and
a processor device operatively coupled to the memory device for storing the program code to
cluster normal historical data of normal cars into a plurality of groups based on a car subsystem to which they belong;
extract (i) features based on membership to the plurality of groups and (ii) feature correlations based on correlation graphs formed from the plurality of groups to retrieve relevant signals and filter signal noise;
train an Auto-Encoder and Auto Decoder (AE&AD) model based on the features and the feature correlations to reconstruct the normal historical data by minimizing reconstruction errors;
reconstruct, using the trained AE&AD model, historical data of specific car fault types with reconstruction errors, normalizing the reconstruction errors, and selecting features of the car faults with a top k large errors as fault signatures; and
reconstruct streaming data of monitored cars using the trained AE&AD model, installed in an edge module of the monitored car that is connected to an online car monitoring system, to determine streaming reconstruction errors, comparing the streaming reconstruction errors with the fault signatures to predict impending known faults and provide alerts for the impending known faults.
20. The computer processing system of claim 19 , wherein the processor device further runs the program code to switch to a backup system to bypass a malfunctioning system responsive to providing an alert for an impending fault.