Machine learning method for anomaly detection in an electrical system
Continuous condition monitoring of an electrical system, and in particular for early fault detection, is provided. The condition monitoring unit is formed with an interface to a memory in which a trained normal state model is stored as a one-class model that has been trained in a training phase with normal state data and represents a normal state of the electrical system. Further, the condition monitoring unit comprises a data interface for continuously acquiring sensor data of the electrical system by means of a set of sensors, an extractor for extracting features from the acquired sensor data, a differentiator for determining deviations of the extracted features from learned features of the normal state model by means of a distance metric (e.g., Euclidean norm, sum norm, maximum norm), a scoring for calculating an anomaly score from the determined deviations, and an output unit for outputting the calculated anomaly score.
1 . A method for continuous condition monitoring of an electrical system, comprising:
in a training phase, training a normal state model as a one-class model with normal state data representing a normal state of the electrical system, wherein a representation or modeling of the normal state is performed via: a bounding-box method, or a k-means method;
in an inference phase, continuously acquiring sensor data of the electrical system using a set of sensors, wherein the set of sensors includes at least one of: voltmeter; ammeter; temperature sensor; speed controller; a sensor for capturing a timestamp; proximity sensor that measures the position of an actuator of the electrical system; linear variable differential transformer (“LVDT”) for displacement measurement of the actuator; limit position switch for detecting when the actuator reaches a certain position; microphone; and/or structure-borne sound pickup;
in a fieldbus node of a unit of the electrical system, extracting features from the acquired sensor data, wherein the unit of the electrical system includes at least one power control and a digital input/output hub, the digital input/output hub being configured to be connected to at least one actuator via a first signal line, and the power control being connected to the digital input/output hub via a second signal line and being configured to control the at least one actuator;
in the fieldbus node of the unit of the electrical system, determining deviations of extracted features from learned features of the normal state model using a distance metric;
in the fieldbus node of the unit of the electrical system, calculating an anomaly score from the determined deviations;
in the fieldbus node of the unit of the electrical system, outputting the calculated anomaly score, wherein the calculated and output anomaly score is used to detect anomalies of the electrical system and wherein the normal state data comprises at least one of: a current strength/torque of a motor, a position, velocity, and/or acceleration of an actuator, microphone/structure-borne sound measurements, force measurements, or one or more signals from one or more electrical components, wherein the one or more electrical components comprises at least one or more cylinders; and
forwarding the anomaly score from the fieldbus node to selected other network participants via a TCP/IP based network protocol.
2 . The method of claim 1 , wherein the normal state model is a statistical model and/or a machine learning model.
3 . The method according to claim 1 , wherein the anomaly score is forwarded to selected other network participants via the TCP/IP-based network protocol, via one of an MQTT protocol or an OPC UA protocol.
4 . The method according to claim 1 , in which a productivity score is determined, when process cycles are automatically detected in order to evaluate how a cycle duration develops over a longer time horizon.
5 . The method according to claim 1 , in which a normalization function, comprising a sigmoid function, is applied to the determined deviations and/or wherein an inflection point and/or a slope of the sigmoid function can be parameterized and/or wherein the sigmoid function is linearly rescaled in the training phase so that a graphical representation of the anomaly score is continuous.
6 . The method according to claim 1 , in which the method is controlled via meta-parameters, wherein the meta-parameters comprise a parameterization of the model, comprising at least a determination of the number of k-means centers and/or a number of bounding boxes and/or a calculation rule for the boundaries of the bounding boxes, and/or a weighting of extracted features and/or further parameters for feature extraction.
7 . The method according to claim 1 , wherein the normal state data in the training phase and the productive data, comprising at least sensor data, in the inference phase are preprocessed using the same preprocessing methods.
8 . The method according to claim 7 , wherein the preprocessing methods comprise an execution of a pattern recognition algorithm on the sensor data and on the normal state data to detect recurring patterns in the sensor data representing process cycles and wherein the detected process cycles are used as parameterization of a time window and/or wherein a result of the pattern recognition algorithm is used to calculate time windows in which the feature extraction is executed.
9 . The method according to claim 7 , wherein the pattern recognition algorithm comprises auto-correlation.
10 . The method according to claim 1 , comprising a dimensionality reduction method, and wherein the dimensionality reduction method is applied to the raw data and/or to the extracted features in a data preprocessing step.
11 . The method according to claim 1 , wherein the calculated anomaly score is subjected to a low-pass filter, the low-pass filter being parameterizable.
12 . The method according to claim 1 , in which sensitivity parameters are detected on an input field of a user interface, the sensitivity parameters characterizing under which conditions comprising at least how quickly differences between the extracted features and the learned features are processed as deviations.
13 . The method according to claim 1 , wherein the extracted features comprise statistical characteristics and comprise mean values, minima, maxima, differences, quantiles, quartiles, skewness and/or kurtosis of the sensor data and/or their derivatives, characteristics of the frequency analysis or other selected characteristics over time.
14 . The method according to claim 1 , wherein the method, after acquiring the sensor data, executes a preprocessing algorithm on the acquired sensor data that transforms the data into a different format and/or filters out outlier data.
15 . A condition monitoring unit for continuous condition monitoring of an electrical system for early fault detection, the condition monitoring unit being designed to carry out the method of claim 1 , having:
an interface to a memory in which a trained normal state model is stored as a one-class model that has been trained in a training phase with normal state data and represents a normal state of the electrical system;
a data interface for continuously acquiring sensor data of the electrical system by means of a set of sensors;
an extractor for extracting features from the acquired sensor data;
a differentiator for determining deviations of the extracted features from learned features of the normal state model using a distance metric;
a scoring unit for calculating an anomaly score from the determined deviations; and
an output unit for outputting the calculated anomaly score.
16 . At least one non-transitory computer readable storage medium comprising instructions which, when the instructions are executed, cause one or more processors to execute the method of claim 1 .