A mechatronic system fault diagnosis system based on deep learning

Abstract

The present invention relates to the technical field of deep learning algorithms, and provides a mechatronic system fault diagnosis system based on deep learning. The data acquisition and preprocessing module acquires the operating data of the mechatronic system in real time, and sets a dynamic window length, cleaning, noise reduction and standardization processing for it; the interpretable deep learning diagnosis module uses causal gating and a dual-branch network to perform feature screening and decoupling learning, extract fault-related features, and generate a diagnosis result containing a causal path and anomaly prompts; the physical constraint fusion module obtains the physical principle data of the mechatronic system and the parameter data of the mechatronic system under normal working conditions, and physically constrains the interpretable deep learning diagnosis module through the physical principles of the mechatronic system; the present invention realizes the causal explicit expression and physical logic consistency guarantee of fault diagnosis through causal feature screening, dynamic causal mask generation and path extraction, fusion of multi-physical field law modeling and physical constraint injection, and combined with time series instantaneous causal analysis.

Description

Technical Field

[0001] The present invention relates to the technical field of deep learning algorithms, and in particular to a mechatronic system fault diagnosis system based on deep learning. Background Art

[0002] Electromechanical system fault diagnosis technology has gone through the development stages from rule-based and signal processing to shallow machine learning. Early methods relied on expert experience to build a rule base, such as expert systems, or extracted fault features through signal processing technologies such as Fourier transform and wavelet analysis.

[0003] As the amount of data increases, shallow machine learning methods such as support vector machines and random forests have been introduced, but they still require manual feature design and have limited ability to handle complex nonlinear problems.

[0004] Deep learning automatically extracts data features through multi-layer neural networks, overcoming the limitations of traditional methods that rely on manual feature engineering. For example, convolutional neural networks excel at processing images and time series data and are often used for analyzing motor vibration signals. Recurrent neural networks capture the dynamic characteristics of time series and are used for dynamic fault diagnosis in power systems. Autoencoder networks extract implicit features through unsupervised learning and are used for noise reduction and feature extraction in complex equipment.

[0005] However, the prior art still has the following disadvantages:

[0006] 1. Black box characteristics. Existing deep learning models are insufficient in explaining the causes of faults. For example, the existing deep learning models are not clear enough to determine whether "air conditioning shutdown" is due to sensor failure or compressor damage. This makes it difficult for China Railway's operation and maintenance personnel to make quick decisions. They still need to rely on manual re-inspection, which delays emergency repairs and affects the safe operation of high-speed railway stations.

[0007] 2. Existing deep learning models are mostly data-driven and lack the integration of the physical principles of electromechanical systems, such as mechanical dynamics and thermodynamics. This can lead to diagnostic results that violate common sense. For example, a model might determine that a sudden drop in bearing temperature is a fault, but in actual operating conditions, such a phenomenon is physically impossible without the intervention of a cooling system. Summary of the Invention

[0008] In order to solve the above technical problems, the present invention provides a mechatronic system fault diagnosis system based on deep learning, which realizes the explicit expression of causal relationship and the guarantee of physical logic consistency of fault diagnosis through causal feature screening decoupling, dynamic causal mask generation and path extraction, integration of multi-physics field law modeling and residual constraint injection, and combination of temporal instantaneous causal analysis and feature purification, so as to solve the problems in the prior art.

[0009] The technical solutions of the present invention are as follows:

[0010] A deep learning-based electromechanical system fault diagnosis system, comprising:

[0011] Data acquisition and preprocessing module, which is used to collect historical and real-time operation data of the electromechanical system, and set dynamic window length, cleaning, noise reduction and standardization processing;

[0012] An interpretable deep learning diagnostic module, which is connected to the data acquisition and preprocessing module, uses causal gating and a dual-branch network to perform feature screening and decoupling learning, extracts fault-related features, and ultimately generates a diagnostic result containing a causal path and anomaly prompts, achieving fault diagnosis while outputting the diagnostic basis;

[0013] A physical constraint fusion module, which is connected to the data of the interpretable deep learning diagnosis module to obtain the physical principle data of the electromechanical system and the parameter data of the electromechanical system under normal operation. The physical constraint fusion module is physically constrained by the physical principles of the electromechanical system to ensure that the diagnosis results conform to the laws of physics.

[0014] A decision interaction module is connected to the data of the explainable deep learning diagnosis module, and is used to receive the results and explanation information output by the explainable deep learning diagnosis module and present them to the operation and maintenance personnel, providing an interactive interface, supporting manual review and feedback, and optimizing the diagnosis system.

[0015] Preferably, the data acquisition and preprocessing module includes a data acquisition unit and a data preprocessing unit;

[0016] The data acquisition unit includes a vibration sensor, a temperature sensor, a current sensor, a voltage sensor and a humidity sensor, which are used to collect real-time operation data of the electromechanical system. At the same time, the data acquisition unit obtains historical electromechanical system operation data and fault data through the electromechanical system historical operation log;

[0017] The data preprocessing unit uses sliding window technology to divide time series data, sets dynamic window length according to different device characteristics, uses adaptive filtering algorithm to eliminate noise interference in the data, and combines normalization method to unify data scale to ensure data quality and consistency.

[0018] Preferably, the explainable deep learning diagnosis module includes a feature calculation unit and an explainable diagnosis unit;

[0019] The feature calculation unit generates a sparse causal graph by fusing time series and instantaneous causal analysis, and combines causal gating with a dual-branch network to achieve feature screening and decoupling learning, thereby effectively extracting fault-related features.

[0020] The explainable diagnosis unit outputs the fault probability distribution based on the fusion features and dynamic causal mask, generates a contribution heat map, identifies unknown anomalies through cluster analysis, and finally generates a diagnosis report containing causal paths and anomaly prompts.

[0021] Preferably, the feature calculation unit includes an offline calculation subunit and an online calculation subunit;

[0022] The offline calculation subunit receives the input multi-sensor time series data matrix and fault label y, where represents the set of real numbers, T×D represents the dimension of the matrix, T is the total length of the time series, and D is the sensor dimension;

[0023] Adopting the time domain causal discovery algorithm, integrating the temporal causal and instantaneous causal analysis, each sensor data , using a 1D causal convolution kernel sliding window to calculate the fault label based on the binary fault label The causal strength score :

[0024] ;

[0025] in, Representing sensor data Fault label The causal strength score of is the temporal causal term, is the instantaneous causal term, It represents the balance factor, which is used to control the weight ratio of the temporal causal term and the instantaneous causal term, and is verified by cross selection;

[0026] Temporal causal terms The specific calculation is:

[0027] ;

[0028] Among them, the smaller the score, the stronger the causality, l represents the length of the causal convolution kernel, that is, the size of the time window, Represents the weight parameter of the convolution kernel k that can be learned, which is used to weight the historical signal Indicates that at time step The fault label is 0 or 1, where 0 indicates no fault and 1 indicates fault, i indicates the i-th sensor and j indicates the j-th fault;

[0029] Instantaneous causal term Representing sensor data With fault label The mutual information is used to quantify the instantaneous statistical dependence between the two, which is independent of the time order. The calculation formula is:

[0030] ;

[0031] Among them, x represents the discretized value of a sensor signal, and y represents a value state of the fault label, that is, 0 or 1; Representing sensor data At x, and fault The joint probability of being in y, Representing sensor data The marginal probability of being at x, Indicates a fault tag In state The marginal probability of

[0032] Calculate the score Then, the significance test, i.e., the permutation test, is used to filter out false associations and generate a sparse causal graph G, where the nodes are sensor i and fault j, and the edges are significant causal links. ;

[0033] Then convert the causal graph G into a binary mask matrix , where C is the number of fault categories, Indicates that sensor i has a causal contribution to fault j; then, at every specified training cycle, the causal graph is recalculated and M is updated to adapt to changes in data distribution;

[0034] The online computing subunit receives input, sensor data And the mask matrix M, through the causal gating layer, performs channel-level filtering on the original features:

[0035] ;

[0036] in, Represents the Hadamard product, forcing the weights of irrelevant channels to zero. At the same time, only the sensor signals causally related to the current fault category are retained for subsequent calculations;

[0037] Then we proceed to decouple feature learning, using a dual-branch network structure:

[0038] Causal branching, processing , extract the temporal causal feature H through causal convolution;

[0039] The non-causal branch processes data X and extracts global features through ordinary CNN ;

[0040] Output fusion features .

[0041] Preferably, the explainable diagnosis unit includes an inference diagnosis subunit and a report generation subunit;

[0042] The reasoning diagnosis subunit outputs the fault probability distribution based on the fusion feature F and the dynamic causal mask M ;

[0043] ;

[0044] in, is a weight matrix whose dimension is the number of fault categories multiplied by the dimension of the fusion feature F. b represents the bias term. Each category corresponds to a scalar bias, which is obtained by training based on the operation data and fault data of the historical electromechanical system. The gradient of the causal branch is traced significantly to generate the fault contribution heat map A, that is, the fault Causal characteristics The sum of the time dimension gradients;

[0045] ;

[0046] Then, for the non-causal branch features Perform K-means clustering. If the current sample deviates from the normal cluster center by more than a threshold, the sample is marked with a non-causal anomaly flag and a non-causal anomaly warning is triggered.

[0047] The report generating subunit is based on the failure probability , the contribution heat map A and non-causal anomaly flags are combined with the dynamic causal graph G to extract the top-K causal paths of fault category j. If a non-causal anomaly warning is triggered, an additional prompt is added: "Unverified feature pattern detected, manual verification of equipment anomaly recommended";

[0048] Preferably, the physical constraint fusion module includes a physical modeling unit, a physical optimization unit and a physical verification unit;

[0049] The physical modeling unit symbolizes physical laws by integrating the multi-physics principles of electromechanical systems, generates differentiable computational graphs, uses PINN to process complex physical relationships, and implements cross-field coupling through tensor operations. It ultimately outputs a DPCU library that supports gradient backpropagation, laying the foundation for subsequent calculations.

[0050] The physical optimization unit injects dynamic physical constraints in real time during model training, estimates the physical state based on raw sensor data and the DPCU Library, calculates the absolute residuals between actual and theoretical physical quantities, and dynamically adjusts the constraint weights in combination with the sliding window and KL divergence to ensure that model training complies with physical laws.

[0051] The physical verification unit verifies the physical feasibility of the prediction results based on the fault type and absolute residual predicted by the model, and determines the validity of the results by searching for the violated physical rule set. For infeasible results, it selects high-probability candidate faults to correct the prediction. If the correction is not possible, it triggers an alarm and traces the suspicious sensor data to provide a basis for system maintenance.

[0052] Preferably, the physical modeling unit inputs the multi-physics principle of the electromechanical system and converts the physical laws into symbolic expressions, and uses the SymPy integration of the automatic differentiation tool PyTorch to generate a differentiable computational graph;

[0053] For complex physical relationships that are difficult to express explicitly, a physical information neural network (PINN) is used for proxy modeling to ensure differentiability, build a field coupling relationship diagram, define the interaction rules of physical quantities, and implement cross-field calculations through tensor operations. Finally, the differentiable physical computing unit library (DPCU Library) is output and supports gradient backpropagation.

[0054] Preferably, the physical optimization unit performs dynamic physical constraint injection during online training of the feature calculation unit;

[0055] First, input the raw sensor data X and DPCU Library to estimate the physical state, and use DPCU to calculate the derivative physical quantity from X. At the same time, theoretical physical quantities are calculated based on physical models , where z represents the number of types of physical quantities;

[0056] Physical residual generation: Calculate the absolute residual between actual and theoretical:

[0057] ;

[0058] And statistically analyze the residual distribution through sliding windows , adaptive constraint relaxation, design dynamic constraint weights:

[0059] ;

[0060] Among them, KL represents KL divergence, which is used to measure the difference between two distributions. represents the scaling factor, controlling the difference sensitivity, is the Sigmoid function, The residual distribution of normal working condition is the more the residual distribution deviates from the normal condition. The larger it is, the stricter the physical constraints are;

[0061] The actual application process is as follows:

[0062] Sliding window statistics, calculate the distribution of the residuals of each physical quantity in the current window every specified time window , KL divergence calculation, With pre-existing Calculate the KL divergence and convert the KL divergence into weights through the Sigmoid function , the greater the difference, The closer it is to 1, the stronger the physical constraint;

[0063] Preferably, the physical verification unit predicts the absolute residual R when the fault types y and t are predicted based on the model;

[0064] Physical feasibility check: according to the fault type y, retrieve the set of physical rules that must be violated ;

[0065] If exists If it is not satisfied, the prediction result y is judged to be physically infeasible;

[0066] Multi-hypothesis correction, enabling physics-guided candidate set search: from failure probability distribution The top three high screening failure probability Type of fault;

[0067] If no candidate meets the requirement, an unknown fault alarm is triggered, prompting manual intervention, physical parameter tracing, and reverse tracing of the contradictory sensor data for physically unfeasible predictions. A list of suspicious sensors is generated for maintenance reference.

[0068] Preferably, the decision interaction module displays the contribution heat map, causal path diagram and original signal comparison, as well as the additional prompts generated in the report generation subunit by providing a web interface.

[0069] Compared with the prior art, the present invention has the following beneficial effects:

[0070] 1. The present invention achieves the explicit presentation of the causal relationship between sensor signals and fault types in the fault diagnosis process through the decoupling of causal gating and feature screening of the dual-branch network in the interpretable deep learning diagnosis module, the generation of fault contribution heat maps and causal path extraction using dynamic causal masks, and achieves a clear distinction between sensor faults and specific failure modes in fault scenarios. This allows operation and maintenance personnel to quickly locate the root cause of the fault based on the causal path and contribution heat map, avoiding delays in manual re-inspection and improving the emergency repair efficiency and operational safety of the electromechanical system of high-speed railway stations.

[0071] 2. The present invention converts multi-physical field laws into differentiable calculation graphs through the physical modeling unit in the physical constraint fusion module, dynamically injects physical residual constraints into the physical optimization unit, and verifies the prediction results based on physical rules by the physical verification unit. It realizes the deep coupling of physical laws such as mechanical dynamics and thermodynamics with deep learning models, and achieves automatic identification and correction of diagnostic results that violate physical common sense, ensuring that the diagnostic results are consistent with the physical feasibility of actual working conditions and avoiding operation and maintenance errors caused by model misjudgment.

[0072] 3. The present invention generates a sparse causal graph through timing and instantaneous causal analysis in the feature calculation unit, filters non-causal correlation features through the causal gating layer, and decouples timing causal features and global features through a dual-branch network. It realizes the targeted screening and efficient extraction of fault-related features in multi-sensor data such as vibration and temperature, achieves the feature purification effect of eliminating noise interference and irrelevant signals, improves the model's ability to capture early fault features, and significantly enhances the accuracy and robustness of electromechanical system fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 It is a schematic diagram of the workflow connection between the units of the present invention;

[0074] Figure 2 Schematic diagram of the calculation process of the interpretable deep learning diagnostic module of the present invention;

[0075] Figure 3 It is a schematic diagram of the calculation flow of the physical constraint fusion module of the present invention. DETAILED DESCRIPTION

[0076] The following embodiments of the present invention are described in further detail with reference to the accompanying drawings and examples. The following examples are used to illustrate the present invention but are not intended to limit the scope of the present invention.

[0077] The present invention provides a deep learning-based electromechanical system fault diagnosis system, comprising:

[0078] Data acquisition and preprocessing module, which is used to collect historical and real-time operation data of the electromechanical system, and set dynamic window length, cleaning, noise reduction and standardization processing;

[0079] An interpretable deep learning diagnostic module, which is connected to the data acquisition and preprocessing module, uses causal gating and a dual-branch network to perform feature screening and decoupling learning, extracts fault-related features, and ultimately generates a diagnostic result containing a causal path and anomaly prompts, achieving fault diagnosis while outputting the diagnostic basis;

[0080] A physical constraint fusion module, which is connected to the data of the interpretable deep learning diagnosis module to obtain the physical principle data of the electromechanical system and the parameter data of the electromechanical system under normal operation. The physical constraint fusion module is physically constrained by the physical principles of the electromechanical system to ensure that the diagnosis results conform to the laws of physics.

[0081] A decision interaction module is connected to the data of the explainable deep learning diagnosis module, and is used to receive the results and explanation information output by the explainable deep learning diagnosis module and present them to the operation and maintenance personnel, providing an interactive interface, supporting manual review and feedback, and optimizing the diagnosis system. Example 1:

[0082] like Figure 1-Figure 3 As shown, in this embodiment, various key electromechanical systems such as ventilation, refrigeration, and power transmission in a hub-level high-speed railway station are operating under high load for a long time. During peak holiday passenger flow periods, when multiple refrigeration units are operating in conjunction to adjust the indoor temperature and humidity, the system monitoring interface suddenly displays abnormal fluctuations in equipment operating parameters in a certain area. At the same time, the equipment operating noise and vibration monitoring module simultaneously captures non-periodic abnormal noises. Operation and maintenance personnel initially determine that this is an equipment failure. However, traditional monitoring methods cannot quickly distinguish whether the failure is caused by sensor signal interference, control logic anomalies, or mechanical component wear. Therefore, the present invention is introduced to avoid large-scale equipment downtime that affects passenger travel.

[0083] The data acquisition unit includes a vibration sensor, a temperature sensor, a current sensor, a voltage sensor and a humidity sensor, which are used to collect real-time operation data of the electromechanical system. At the same time, the data acquisition unit obtains historical electromechanical system operation data and fault data through the electromechanical system historical operation log;

[0084] The data preprocessing unit uses sliding window technology to divide time series data, sets dynamic window length according to different device characteristics, applies adaptive filtering algorithm to eliminate noise interference in the data, and combines normalization method to unify data scale to ensure data quality and consistency.

[0085] The feature calculation unit generates a sparse causal graph by fusing time series and instantaneous causal analysis. It then combines causal gating with a dual-branch network to implement feature screening and decoupled learning, effectively extracting fault-related features.

[0086] The feature calculation unit includes an offline calculation subunit and an online calculation subunit;

[0087] The offline computing subunit receives the input multi-sensor time series data matrix and fault label y, where represents the set of real numbers, T×D represents the dimension of the matrix, T is the total length of the time series, and D is the sensor dimension;

[0088] Adopting the time domain causal discovery algorithm, integrating the temporal causal and instantaneous causal analysis, each sensor data , using a 1D causal convolution kernel sliding window to calculate the fault label based on the binary fault label The causal strength score :

[0089] ;

[0090] in, Representing sensor data Fault label The causal strength score of is the temporal causal term, is the instantaneous causal term, It represents the balance factor, which is used to control the weight ratio of the temporal causal term and the instantaneous causal term, and is verified by cross selection;

[0091] Temporal causal terms The specific calculation is:

[0092] ;

[0093] Among them, the smaller the score, the stronger the causality, l represents the length of the causal convolution kernel, that is, the size of the time window, Represents the weight parameter of the convolution kernel k that can be learned, which is used to weight the historical signal Indicates that at time step The fault label is 0 or 1, where 0 indicates no fault and 1 indicates fault, i indicates the i-th sensor and j indicates the j-th fault;

[0094] Instantaneous causal term Representing sensor data With fault label The mutual information is used to quantify the instantaneous statistical dependence between the two, which is independent of the time order. The calculation formula is:

[0095] ;

[0096] Among them, x represents the discretized value of a sensor signal, and y represents a value state of the fault label, that is, 0 or 1; Representing sensor data At x, and fault The joint probability of being in y, Representing sensor data The marginal probability of being at x, Indicates a fault tag In state The marginal probability of

[0097] Calculate the score Then, the significance test, i.e., the permutation test, is used to filter out false associations and generate a sparse causal graph G, where the nodes are sensor i and fault j, and the edges are significant causal links. ;

[0098] Then convert the causal graph G into a binary mask matrix , where C is the number of fault categories, Indicates that sensor i has a causal contribution to fault j; then, at every specified training cycle, the causal graph is recalculated and M is updated to adapt to changes in data distribution;

[0099] The online computing subunit receives input, sensor data And the mask matrix M, through the causal gating layer, performs channel-level filtering on the original features:

[0100] ;

[0101] in, Represents the Hadamard product, forcing the weights of irrelevant channels to zero. At the same time, only the sensor signals causally related to the current fault category are retained for subsequent calculations;

[0102] Then we proceed to decouple feature learning, using a dual-branch network structure:

[0103] Causal branching, processing , extract the temporal causal feature H through causal convolution;

[0104] The non-causal branch processes data X and extracts global features through ordinary CNN ;

[0105] Output fusion features .

[0106] By generating a sparse causal graph through timing and instantaneous causal analysis in the feature calculation unit, filtering non-causal correlation features through the causal gating layer, and decoupling the timing causal features and global features through a dual-branch network, the targeted screening and efficient extraction of fault-related features in multi-sensor data such as vibration and temperature are achieved, achieving the feature purification effect of eliminating noise interference and irrelevant signals, improving the model's ability to capture early fault features, and significantly enhancing the accuracy and robustness of electromechanical system fault diagnosis.

[0107] The interpretable diagnosis unit outputs the fault probability distribution based on the fusion feature and the dynamic causal mask, generates a contribution heat map, and identifies unknown anomalies through cluster analysis, and finally generates a diagnosis report containing causal paths and anomaly prompts. The interpretable diagnosis unit includes a reasoning diagnosis subunit and a report generation subunit. The reasoning diagnosis subunit outputs the fault probability distribution based on the fusion feature F and the dynamic causal mask M. ;

[0108] ;

[0109] in, is a weight matrix whose dimension is the number of fault categories multiplied by the dimension of the fusion feature F. b represents the bias term. Each category corresponds to a scalar bias, which is obtained by training based on the operation data and fault data of the historical electromechanical system. The gradient of the causal branch is traced significantly to generate the fault contribution heat map A, that is, the fault Causal characteristics The sum of the time dimension gradients;

[0110] ;

[0111] Then, for the non-causal branch features Perform K-means clustering. If the current sample deviates from the normal cluster center by more than a threshold, the sample is marked with a non-causal anomaly flag and a non-causal anomaly warning is triggered.

[0112] Report generation subunit based on failure probability , the contribution heat map A and non-causal anomaly flags are combined with the dynamic causal graph G to extract the top-K causal paths of fault category j. If a non-causal anomaly warning is triggered, an additional prompt is added: "Unverified feature pattern detected, manual verification of equipment anomaly recommended";

[0113] By decoupling the feature screening of causal gating and dual-branch networks in the interpretable deep learning diagnosis module, generating fault contribution heat maps and extracting causal paths using dynamic causal masks, the causal relationship between sensor signals and fault types is made explicit during the fault diagnosis process, achieving a clear distinction between sensor faults and specific failure modes in fault scenarios. Operation and maintenance personnel can quickly locate the root cause of the fault based on the causal path and contribution heat map, avoiding delays in manual re-inspection and improving the repair efficiency and operational safety of the electromechanical system of high-speed railway stations.

[0114] Example 2:

[0115] like Figure 1-Figure 3As shown, in this embodiment, the high-speed railway station building in Example 1 was hit by heavy rain for several consecutive days at a certain time, and outdoor rainwater flowed back into the equipment interlayer, causing the humidity in some electromechanical equipment compartments to rise sharply. The operation and maintenance personnel found that the bearing temperature sensors of multiple water pump units continued to feedback abnormalities, while the vibration sensor data did not change significantly, and the equipment operating power curve also remained stable. At this time, the model's preliminary diagnosis prompted a "potential mechanical failure risk", but on-site operation and maintenance experience shows that: under the working conditions where the active cooling system is not enabled and the equipment load does not change suddenly, a sharp drop in bearing temperature in a short period of time violates the basic laws of thermodynamics. There may be distortion of the sensor signal due to moisture or a conflict in the physical model prediction. It is necessary to verify whether the diagnostic results are consistent with the equipment dynamics principle to avoid the risk of excessive maintenance or missed inspections due to misjudgment.

[0116] The physical modeling unit symbolizes physical laws by integrating the multi-physics principles of electromechanical systems, generates differentiable computational graphs, uses PINN to process complex physical relationships, and implements cross-field coupling through tensor operations. It ultimately outputs a DPCU library that supports gradient backpropagation, laying the foundation for subsequent calculations.

[0117] The physical modeling unit inputs the multi-physics principles of electromechanical systems, such as the law of conservation of energy, and converts these physical laws into symbolic expressions. It uses the SymPy integration of the automatic differentiation tool PyTorch to generate differentiable computational graphs.

[0118] For complex physical relationships that are difficult to express explicitly, a physical information neural network (PINN) is used for proxy modeling to ensure differentiability, build a field coupling relationship diagram, define the interaction rules of physical quantities, and implement cross-field calculations through tensor operations. Finally, the differentiable physical computing unit library (DPCU Library) is output and supports gradient backpropagation.

[0119] The physical optimization unit performs dynamic physical constraint injection during the online training of the feature calculation unit;

[0120] First, input the raw sensor data X and DPCU Library to estimate the physical state, and use DPCU to calculate the derivative physical quantity from X. At the same time, theoretical physical quantities are calculated based on physical models , where z represents the number of types of physical quantities;

[0121] Physical residual generation: Calculate the absolute residual between actual and theoretical:

[0122] ;

[0123] And statistically analyze the residual distribution through sliding windows , adaptive constraint relaxation, design dynamic constraint weights:

[0124] ;

[0125] Among them, KL represents KL divergence, which is used to measure the difference between two distributions. represents the scaling factor, controlling the difference sensitivity, is the Sigmoid function, The residual distribution of normal working condition is the more the residual distribution deviates from the normal condition. The larger it is, the stricter the physical constraints are;

[0126] The actual application process is as follows:

[0127] Sliding window statistics, calculate the distribution of the residuals of each physical quantity in the current window every specified time window , KL divergence calculation, With pre-existing Calculate the KL divergence and convert the KL divergence into weights through the Sigmoid function , the greater the difference, The closer it is to 1, the stronger the physical constraint;

[0128] The absolute residual R when the physical verification unit predicts the fault types y and t based on the model;

[0129] Physical feasibility check: according to the fault type y, retrieve the set of physical rules that must be violated ;

[0130] If exists If it is not satisfied, the prediction result y is judged to be physically infeasible;

[0131] Multi-hypothesis correction, enabling physics-guided candidate set search: from failure probability distribution The top three high screening failure probability Type of fault;

[0132] If no candidate meets the requirement, an unknown fault alarm is triggered, prompting manual intervention, physical parameter tracing, and reverse tracing of the contradictory sensor data for physically unfeasible predictions. A list of suspicious sensors is generated for maintenance reference.

[0133] Through the physical modeling unit in the physical constraint fusion module, the multi-physical field laws are converted into differentiable calculation graphs, the physical optimization unit dynamically injects physical residual constraints, and the physical verification unit verifies the prediction results based on physical rules. This realizes the deep coupling of physical laws such as mechanical dynamics and thermodynamics with deep learning models, and achieves automatic identification and correction of diagnostic results that violate physical common sense, ensuring that the diagnostic results are consistent with the physical feasibility of actual working conditions and avoiding operation and maintenance errors caused by model misjudgment.

[0134] The embodiments of the present invention are provided for the purpose of illustration and description. Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations of the present invention. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A deep learning-based electromechanical system fault diagnosis system, characterized in that: include: Data acquisition and preprocessing module, which is used to collect historical and real-time operation data of the electromechanical system, and set dynamic window length, cleaning, noise reduction and standardization processing; An interpretable deep learning diagnostic module, which is connected to the data acquisition and preprocessing module, uses causal gating and a dual-branch network to perform feature screening and decoupling learning, extracts fault-related features, and ultimately generates a diagnostic result containing a causal path and anomaly prompts, achieving fault diagnosis while outputting the diagnostic basis; The explainable deep learning diagnosis module includes a feature calculation unit and an explainable diagnosis unit; The feature calculation unit includes an offline calculation subunit and an online calculation subunit; The online computing subunit receives input, sensor data And the mask matrix M, through the causal gating layer, performs channel-level filtering on the original features: ； in, Represents the Hadamard product, forcing the weights of irrelevant channels to zero. At the same time, only the sensor signals causally related to the current fault category are retained for subsequent calculations; Then we proceed to decouple feature learning, using a dual-branch network structure: Causal branching, processing , extract the temporal causal feature H through causal convolution; The non-causal branch processes data X and extracts global features through ordinary CNN ; Output fusion features ; A physical constraint fusion module, which is connected to the data of the interpretable deep learning diagnosis module to obtain the physical principle data of the electromechanical system and the parameter data of the electromechanical system under normal operation. The physical constraint fusion module is physically constrained by the physical principles of the electromechanical system to ensure that the diagnosis results conform to the laws of physics. A decision interaction module is connected to the data of the explainable deep learning diagnosis module, and is used to receive the results and explanation information output by the explainable deep learning diagnosis module and present them to the operation and maintenance personnel, providing an interactive interface, supporting manual review and feedback, and optimizing the diagnosis system.

2. The electromechanical system fault diagnosis system based on deep learning according to claim 1, characterized in that: The data acquisition and preprocessing module includes a data acquisition unit and a data preprocessing unit; The data acquisition unit includes a vibration sensor, a temperature sensor, a current sensor, a voltage sensor and a humidity sensor, which are used to collect real-time operation data of the electromechanical system. At the same time, the data acquisition unit obtains historical electromechanical system operation data and fault data through the electromechanical system historical operation log; The data preprocessing unit uses sliding window technology to divide time series data, sets dynamic window length according to different device characteristics, uses adaptive filtering algorithm to eliminate noise interference in the data, and combines normalization method to unify data scale to ensure data quality and consistency.

3. The electromechanical system fault diagnosis system based on deep learning according to claim 1, characterized in that: The feature calculation unit generates a sparse causal graph by fusing time series and instantaneous causal analysis, and combines causal gating with a dual-branch network to achieve feature screening and decoupling learning, thereby effectively extracting fault-related features. The explainable diagnosis unit outputs the fault probability distribution based on the fusion features and dynamic causal mask, generates a contribution heat map, identifies unknown anomalies through cluster analysis, and finally generates a diagnosis report containing causal paths and anomaly prompts.

4. The electromechanical system fault diagnosis system based on deep learning according to claim 3, characterized in that: The offline calculation subunit receives the input multi-sensor time series data matrix and fault label y, where represents the set of real numbers, T×D represents the dimension of the matrix, T is the total length of the time series, and D is the sensor dimension; Adopting the time domain causal discovery algorithm, integrating the temporal causal and instantaneous causal analysis, each sensor data , using a 1D causal convolution kernel sliding window to calculate the fault label based on the binary fault label The causal strength score : ； in, Representing sensor data Fault label The causal strength score of is the temporal causal term, is the instantaneous causal term, It represents the balance factor, which is used to control the weight ratio of the temporal causal term and the instantaneous causal term, and is verified by cross selection; Temporal causal terms The specific calculation is: ； Among them, the smaller the score, the stronger the causality, l represents the length of the causal convolution kernel, that is, the size of the time window, Represents the weight parameter of the convolution kernel k that can be learned, which is used to weight the historical signal Indicates that at time step The fault label is 0 or 1, where 0 indicates no fault and 1 indicates fault, i indicates the i-th sensor and j indicates the j-th fault; Instantaneous causal term Representing sensor data With fault label The mutual information is used to quantify the instantaneous statistical dependence between the two, which is independent of the time order. The calculation formula is: ； Among them, x represents the discretized value of a sensor signal, and y represents a value state of the fault label, that is, 0 or 1; Representing sensor data At x, and fault The joint probability of being in y, Representing sensor data The marginal probability of being at x, Indicates a fault tag In state The marginal probability of Calculate the score Then, the significance test, i.e., the permutation test, is used to filter out false associations and generate a sparse causal graph G, where the nodes are sensor i and fault j, and the edges are significant causal links. ; Then convert the causal graph G into a binary mask matrix , where C is the number of fault categories, Indicates that sensor i has a causal contribution to fault j; then, at every specified training cycle, the causal graph is recalculated and M is updated to adapt to changes in data distribution.

5. The electromechanical system fault diagnosis system based on deep learning according to claim 4, characterized in that: The explainable diagnosis unit includes an inference diagnosis subunit and a report generation subunit; The reasoning diagnosis subunit outputs the fault probability distribution based on the fusion feature F and the dynamic causal mask M ; ； in, is a weight matrix whose dimension is the number of fault categories multiplied by the dimension of the fusion feature F. b represents the bias term. Each category corresponds to a scalar bias, which is obtained by training based on the operation data and fault data of the historical electromechanical system. The gradient of the causal branch is traced significantly to generate the fault contribution heat map A, that is, the fault Causal characteristics The sum of the time dimension gradients; ； Then, for the non-causal branch features Perform K-means clustering. If the current sample deviates from the normal cluster center by more than a threshold, the sample is marked with a non-causal anomaly flag and a non-causal anomaly warning is triggered. The report generating subunit is based on the failure probability , the contribution heat map A and non-causal anomaly signs are combined with the dynamic causal graph G to extract the Top-K causal paths of fault category j. If a non-causal anomaly warning is triggered, an additional prompt is added: "Unverified feature pattern detected, manual verification of equipment anomaly is recommended." 6. The electromechanical system fault diagnosis system based on deep learning according to claim 5, characterized in that: The physical constraint fusion module includes a physical modeling unit, a physical optimization unit and a physical verification unit; The physical modeling unit symbolizes physical laws by integrating the multi-physics principles of electromechanical systems, generates differentiable computational graphs, uses PINN to process complex physical relationships, and implements cross-field coupling through tensor operations. It ultimately outputs a DPCU library that supports gradient backpropagation, laying the foundation for subsequent calculations. The physical optimization unit injects dynamic physical constraints in real time during model training, estimates the physical state based on raw sensor data and DPCULibrary, calculates the absolute residuals between actual and theoretical physical quantities, and dynamically adjusts the constraint weights in combination with sliding windows and KL divergence to ensure that model training complies with physical laws. The physical verification unit verifies the physical feasibility of the prediction results based on the fault type and absolute residual predicted by the model, and determines the validity of the results by retrieving the violated physical rule set; for infeasible results, it screens high-probability candidate fault correction predictions. If correction is not possible, an alarm is triggered and suspicious sensor data is traced back to provide a basis for system maintenance.

7. The electromechanical system fault diagnosis system based on deep learning according to claim 6, characterized in that: The physical modeling unit inputs the multi-physics principle of the electromechanical system and converts the physical laws into symbolic expressions, using the SymPy integration of the automatic differentiation tool PyTorch to generate a differentiable computational graph; For complex physical relationships that are difficult to express explicitly, the physical information neural network PINN is used for proxy modeling to ensure differentiability, construct a field coupling relationship diagram, define the interaction rules of physical quantities, realize cross-field calculations through tensor operations, and finally output the differentiable physical computing unit library DPCU Library, and support gradient backpropagation.

8. The electromechanical system fault diagnosis system based on deep learning according to claim 7, characterized in that: The physical optimization unit performs dynamic physical constraint injection during online training of the feature calculation unit; First, input the raw sensor data X and DPCU Library to estimate the physical state, and use DPCU to calculate the derivative physical quantity from X. At the same time, theoretical physical quantities are calculated based on physical models , where z represents the number of types of physical quantities; Physical residual generation: Calculate the absolute residual between actual and theoretical: ； And statistically analyze the residual distribution through sliding windows , adaptive constraint relaxation, design dynamic constraint weights: ； Among them, KL represents KL divergence, which is used to measure the difference between two distributions. represents the scaling factor, controlling the difference sensitivity, is the Sigmoid function, The residual distribution of normal working condition is the more the residual distribution deviates from the normal condition. The larger it is, the stricter the physical constraints are; The actual application process is as follows: Sliding window statistics, calculate the distribution of the residuals of each physical quantity in the current window every specified time window , KL divergence calculation, With pre-existing Calculate the KL divergence and convert the KL divergence into weights through the Sigmoid function , the greater the difference, The closer it is to 1, the stronger the physical constraint.

9. The electromechanical system fault diagnosis system based on deep learning according to claim 8, characterized in that: The physical verification unit predicts the absolute residual R of the fault types y and t based on the model; Physical feasibility check: according to the fault type y, retrieve the set of physical rules that must be violated ; If exists If it is not satisfied, the prediction result y is judged to be physically infeasible; Multi-hypothesis correction, enabling physics-guided candidate set search: from failure probability distribution The top three high screening failure probability Type of fault; If no candidate meets the requirement, an unknown fault alarm is triggered, prompting manual intervention, physical parameter tracing, physical infeasibility prediction, reverse tracing of contradictory sensor data, and generating a list of suspicious sensors for maintenance reference.

10. The electromechanical system fault diagnosis system based on deep learning according to claim 5, characterized in that: The decision interaction module displays the contribution heat map, causal path diagram and original signal comparison, as well as the additional prompts generated in the report generation subunit by providing a web interface.

Images