CN-122020491-A - Digital twinning-based miniature production line real-time monitoring and fault prediction method and system

Abstract

The application provides a digital twin-based miniature production line real-time monitoring and fault prediction method and system, which relate to the technical field of fault prediction and comprise the steps of acquiring reference data, real-time data and continuous waveform signals of the miniature production line, and inputting the real-time data and the reference data into a digital twin model for comparison to obtain structural change data; and finally, screening an abnormal event queue through an isolated forest algorithm, carrying out causal verification, and feeding back a verification report to the digital twin model. The application can realize the positioning and prediction of the fault source with high reliability under the microsecond fault propagation scene.

Inventors

• Jin Yueran
• YIN JINLIANG

Assignees

• 天津理工大学

Dates

Publication Date

20260512

Application Date

20260414

Claims (10)

1. 1. The digital twinning-based miniature production line real-time monitoring and fault prediction method is characterized by comprising the following steps of: acquiring reference data of a miniature production line at the checking and accepting time, real-time data of the miniature production line in an operation period and continuous waveform signals of a mechanical framework; inputting the real-time data and the reference data into a digital twin model for comparison to obtain structure change data; Performing sparse transformation on the continuous waveform signals by adopting a sparse representation model to generate a sparse event sequence, and performing plastic learning on the sparse event sequence according to the structural change data to obtain a plurality of suspected fault marks; And screening the abnormal modes of all the suspected fault marks through an isolated forest algorithm to obtain an abnormal event queue, performing causal verification on the abnormal event queue to obtain a verification report, and feeding the verification report back to the digital twin model.
2. 2. The method of claim 1, wherein said performing causal verification on the abnormal event queue to obtain a verification report comprises: Taking mechanical positions corresponding to each suspected fault mark in the abnormal event queue as nodes, and taking the sequence of pulse arrival time among the mechanical positions as directed edges to construct a causal structure diagram consisting of a plurality of nodes and a plurality of directed edges; Executing intervention operation on each node in the causal structure chart by adopting a counterfactual reasoning method, and generating a counterfactual sample corresponding to each node; Inputting the counterfactual samples and the original samples in the abnormal event queue into a twin network, calculating the similarity distance between the corresponding counterfactual samples and the original samples through the twin network, and calculating the contribution factor corresponding to each node according to the similarity distance; Calculating the fault confidence coefficient of each mechanical position according to all the contribution coefficient factors and the structural change data corresponding to each mechanical position, and marking the mechanical position of which the fault confidence coefficient exceeds a preset confidence threshold as a fault source position; and generating a verification report containing the fault source position, the fault confidence and the fault propagation path according to the fault source position and the corresponding downstream nodes in the causal structure diagram.
3. 3. The method according to claim 2, wherein said inputting the counterfactual samples and the original samples in the abnormal event queue into the twin network, calculating the similarity distances between the corresponding counterfactual samples and the original samples by the twin network, comprises: inputting the inverse fact sample corresponding to each mechanical position to a first feature extraction branch of the twin network to perform full convolution processing to obtain an inverse fact feature vector; Inputting an original sample corresponding to the same mechanical position to a second feature extraction branch of the twin network to perform full convolution processing to obtain an original feature vector, wherein the first feature extraction branch and the second feature extraction branch share network weights; Inputting the inverse fact feature vector and the original feature vector into a calculation branch of the twin network, wherein the calculation branch adopts a cross-correlation calculation method to calculate the similarity distance between the inverse fact feature vector and the original feature vector.
4. 4. The method of claim 1, wherein sparsely transforming the continuous waveform signal using a sparse representation model generates a sparse event sequence comprising: Intercepting a plurality of waveform fragments corresponding to the mechanical positions indicated by the structural change data from the continuous waveform signals, taking all the waveform fragments as training sample sets, and adopting a singular value decomposition algorithm to iteratively learn the training sample sets so as to construct an ultra-complete redundant dictionary; taking the ultra-complete redundant dictionary as a sparse representation model, and adopting an orthogonal matching pursuit algorithm to carry out projection decomposition on the continuous waveform signals acquired subsequently on the sparse representation model to obtain a sparse coefficient sequence; and extracting an atomic index, pulse arrival time and atomic amplitude corresponding to each non-zero projection coefficient from the sparse coefficient sequence, and combining the atomic index, the pulse arrival time and the atomic amplitude into a sparse event sequence.
5. 5. The method of claim 1, wherein the performing plastic learning on the sparse event sequence according to the structural change data to obtain a plurality of suspected fault markers comprises: assigning a cluster of impulse neurons to the mechanical location indicated by the structural change data; Calculating the arrival time difference between the presynaptic pulse and the postsynaptic pulse in each pulse neuron cluster by adopting a pulse time dependent plasticity algorithm, determining a weight adjustment direction of long-time enhancement or long-time inhibition according to the arrival time difference, and updating the synaptic weight of the pulse neuron cluster according to the weight adjustment direction; calculating the attention weight of each mechanical position by adopting an attention mechanism, and modulating the synaptic weight according to the attention weight to obtain a modulated synaptic weight; Comparing the modulated synaptic weight with a preset first excitation threshold, recording a mechanical position corresponding to a current pulse neuron cluster as an excitation position when a comparison result shows that the modulated synaptic weight exceeds the first excitation threshold, and combining the excitation position and the corresponding pulse arrival time as candidate fault marks; performing amplitude compression on the synaptic weights corresponding to the candidate fault marks to obtain normalized synaptic weights, and applying a periodic gating signal to the normalized synaptic weights by adopting a dynamic gating mechanism to obtain synaptic weights after applying signals; and performing secondary comparison on the synaptic weight after the signal is applied and a preset second excitation threshold value, and when the secondary comparison result shows that the synaptic weight after the signal is applied exceeds the second excitation threshold value, recording the mechanical position corresponding to the current pulse neuron cluster as a suspected fault position and combining the suspected fault position and the corresponding pulse arrival time as a suspected fault mark.
6. 6. The method according to claim 1, wherein the performing anomaly pattern screening on all the suspected fault marks by using an orphan forest algorithm to obtain an anomaly event queue includes: Distributing a dynamic weight coefficient for each mechanical position according to the change rate of each mechanical position in the structural change data, and multiplying the dynamic weight coefficient by a synaptic weight value of the suspected fault mark of the corresponding mechanical position after the signal is applied to obtain a weighted suspected fault mark; Arranging all the weighted suspected fault marks according to a time sequence to form a periodic mark sequence of each production period, performing dimension reduction transformation on the periodic mark sequence by adopting a random projection method to obtain a low-dimensional feature vector, and inputting the low-dimensional feature vector into an isolated forest algorithm; Randomly selecting a feature dimension and a segmentation value from each isolated tree, recursively segmenting the low-dimensional feature vector until each low-dimensional feature vector is segmented to a leaf node independently, and recording the path length of each low-dimensional feature vector in each isolated tree; Calculating an average path length according to the path length of each low-dimensional feature vector in all the isolated trees, and calculating an abnormality index of each periodic marker sequence according to the average path length; and reserving the suspected fault marks of which the abnormality indexes exceed a preset abnormality threshold in each production period, and arranging the reserved suspected fault marks according to pulse arrival time to form an abnormal event queue.
7. 7. The method of claim 1, wherein said comparing said real-time data with said reference data input into a digital twin model to obtain structural change data comprises: Extracting peak frequency corresponding to each mechanical position from the real-time data, and performing difference operation on the peak frequency and the reference peak frequency of the corresponding mechanical position in the reference data by a comparison engine of a digital twin model to obtain the frequency offset of each mechanical position; Extracting the amplitude attenuation rate corresponding to each mechanical position from the real-time data, and carrying out ratio operation on the amplitude attenuation rate and the reference attenuation rate of the corresponding mechanical position in the reference data by the comparison engine to obtain the amplitude change coefficient of each mechanical position; the frequency offset of each mechanical position and the amplitude change coefficient are weighted and combined through a calculation engine of the digital twin model to obtain a comprehensive change index of each mechanical position; and arranging all the comprehensive change indexes according to mechanical positions through a combination engine of the digital twin model to form structural change data.
8. 8. A digital twinning-based miniature production line real-time monitoring and fault prediction system is characterized by comprising: The acquisition module is used for acquiring reference data of the micro production line at the acceptance time, real-time data of the micro production line in the operation period and continuous waveform signals of the mechanical skeleton; the comparison module is used for inputting the real-time data and the reference data into a digital twin model for comparison to obtain structure change data; the transformation module is used for carrying out sparse transformation on the continuous waveform signals by adopting a sparse representation model to generate a sparse event sequence, and carrying out plastic learning on the sparse event sequence according to the structural change data to obtain a plurality of suspected fault marks; The verification module is used for screening the abnormal modes of all the suspected fault marks through an isolated forest algorithm to obtain an abnormal event queue, carrying out causal verification on the abnormal event queue to obtain a verification report, and feeding the verification report back to the digital twin model.
9. 9. An electronic device, comprising: A memory for storing a computer program; A processor for implementing the steps of the digital twin based micro-line real time monitoring and fault prediction method as claimed in any one of claims 1 to 7 when executing said computer program.
10. 10. A computer readable storage medium, wherein a computer program is stored in the computer readable storage medium, and when the computer program is executed by a processor, the computer program can implement the digital twin-based micro production line real-time monitoring and fault prediction method as claimed in any one of claims 1 to 7.

Description

Digital twinning-based miniature production line real-time monitoring and fault prediction method and system Technical Field The application relates to the technical field of fault prediction, in particular to a digital twinning-based miniature production line real-time monitoring and fault prediction method and system. Background In the field of miniature high-speed production lines, the real-time monitoring and fault prediction of the running state of equipment are realized, which is the key for guaranteeing the production continuity and the product consistency, and has important engineering application value. At present, the fault diagnosis method based on digital twin is mainly divided into two types, namely, the method is characterized in that real-time operation data are collected and compared with a simulation model, a threshold value judgment or a machine learning model is utilized to identify an abnormal state, and the other type is focused on feature extraction of waveform signals such as vibration and the like and pattern matching is conducted by combining a historical fault library so as to realize fault early warning. However, when the existing method is used for processing a miniature high-speed production line with extremely high fault propagation speed, multi-source heterogeneous data are difficult to effectively fuse, and the dependency on fault samples is strong, so that the fault source and causal association of microsecond propagation cannot be accurately traced under the condition that the fault data are scarce, and the reliability and the interpretability of a prediction result are insufficient. Disclosure of Invention The application aims to provide a digital twinning-based miniature production line real-time monitoring and fault prediction method and system, which are used for solving the problem of low reliability of fault prediction in the prior art. In order to solve the technical problems, in a first aspect, the present application provides a method for real-time monitoring and fault prediction of a micro production line based on digital twinning, comprising: acquiring reference data of a miniature production line at the checking and accepting time, real-time data of the miniature production line in an operation period and continuous waveform signals of a mechanical framework; inputting the real-time data and the reference data into a digital twin model for comparison to obtain structure change data; Performing sparse transformation on the continuous waveform signals by adopting a sparse representation model to generate a sparse event sequence, and performing plastic learning on the sparse event sequence according to the structural change data to obtain a plurality of suspected fault marks; And screening the abnormal modes of all the suspected fault marks through an isolated forest algorithm to obtain an abnormal event queue, performing causal verification on the abnormal event queue to obtain a verification report, and feeding the verification report back to the digital twin model. Optionally, the performing causal verification on the abnormal event queue to obtain a verification report includes: Taking mechanical positions corresponding to each suspected fault mark in the abnormal event queue as nodes, and taking the sequence of pulse arrival time among the mechanical positions as directed edges to construct a causal structure diagram consisting of a plurality of nodes and a plurality of directed edges; Executing intervention operation on each node in the causal structure chart by adopting a counterfactual reasoning method, and generating a counterfactual sample corresponding to each node; Inputting the counterfactual samples and the original samples in the abnormal event queue into a twin network, calculating the similarity distance between the corresponding counterfactual samples and the original samples through the twin network, and calculating the contribution factor corresponding to each node according to the similarity distance; Calculating the fault confidence coefficient of each mechanical position according to all the contribution coefficient factors and the structural change data corresponding to each mechanical position, and marking the mechanical position of which the fault confidence coefficient exceeds a preset confidence threshold as a fault source position; and generating a verification report containing the fault source position, the fault confidence and the fault propagation path according to the fault source position and the corresponding downstream nodes in the causal structure diagram. Optionally, the inputting the counterfactual samples and the original samples in the abnormal event queue into a twin network, calculating, by the twin network, similarity distances between the corresponding counterfactual samples and the original samples, including: inputting the inverse fact sample corresponding to each mechanical position to a first feature extraction bran