A new fault and compound fault collaborative diagnosis method, device and storage medium

Abstract

The present invention discloses a collaborative diagnosis method, device and storage medium for new faults and composite faults, belonging to the field of mechanical manufacturing technology. The method includes: constructing training and test data sets, using acceleration sensors installed on mechanical equipment to collect vibration signal data of mechanical equipment under different working conditions; constructing a training model including a feature extractor, a domain discriminator, a new fault detector and a composite fault classifier; optimizing model parameters, based on a collaborative training strategy, integrating adversarial domain adaptation, open domain adaptation, and pseudo-label learning technology, to construct a multi-task learning framework so that the fault diagnosis model obtains optimal network parameters; collaborative diagnosis of known faults, new faults and composite faults, using a well-trained feature extractor and composite fault classifier to construct a test model, and simultaneously realizing known fault diagnosis, new fault detection and composite fault decoupling tasks. The present invention can simultaneously realize known fault diagnosis, new fault detection and composite fault decoupling.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanical manufacturing, and in particular to a new fault and compound fault collaborative diagnosis method, device and storage medium. Background Art

[0002] Fault diagnosis methods for mechanical equipment play a crucial role in ensuring reliable operation and preventing unexpected loss of life and property due to component failures. In recent years, intelligent fault diagnosis techniques based on deep learning have garnered widespread attention among experts and scholars due to their advantages, such as end-to-end fault diagnosis and independence from expert knowledge in model building. Focusing on practical industrial needs, experts and scholars have developed numerous high-performance intelligent fault diagnosis methods and are committed to promoting their deployment in real-world industrial scenarios.

[0003] In real-world industrial scenarios, machinery and equipment operate under constantly changing operating conditions, making them prone to various types of failures. New and compound faults are two typical examples. When a new fault occurs, intelligent diagnostic models trained on known faults are prone to misdiagnosis, making it difficult to effectively identify new fault types. Compound faults are difficult to identify due to the sheer variety of compound fault types. Furthermore, intelligent diagnostic models trained on known faults are prone to misdiagnosing compound faults as single faults, leading to missed diagnoses.

[0004] Experts and scholars have proposed many effective solutions to the problem of diagnosing new and complex faults in mechanical equipment. For example, for new fault diagnosis, an intelligent fault diagnosis model based on the Open Set Adaptation (OSDA) method has been proposed, which leverages OSDA's superior unknown pattern recognition capabilities to identify new fault types. For complex faults, an intelligent fault diagnosis model based on capsule networks has been proposed, which uses capsule layers to decouple complex faults under incomplete data conditions.

[0005] However, traditional intelligent fault diagnosis methods often focus on a single diagnostic task. When new faults and compound faults occur simultaneously in mechanical equipment, intelligent fault diagnosis models often fail to simultaneously detect new faults and decouple compound faults. Furthermore, the excellent performance of traditional intelligent fault diagnosis models often relies on complete data for training. However, in real-world industrial scenarios, collecting complete labeled fault data is impractical.

[0006] Therefore, the existing technical solutions are not sufficient to completely solve the problems existing in current actual industrial scenarios. It is urgently necessary to design an intelligent diagnosis model that does not rely on complete labeled fault data and can collaboratively detect new faults and decouple complex faults. Summary of the Invention

[0007] In order to solve at least one of the technical problems existing in the prior art to a certain extent, the present invention aims to provide a method, device and storage medium for collaborative diagnosis of new faults and compound faults.

[0008] The technical solution adopted in the present invention is:

[0009] A collaborative diagnosis method for new faults and compound faults includes the following steps:

[0010] Collect vibration signal data of mechanical equipment and obtain training sets and test sets based on the vibration signal data;

[0011] A fault diagnosis model is constructed and trained using a training set. The fault diagnosis model includes a feature extractor, a domain discriminator, a new fault detector, and a composite fault classifier. The feature extractor is used to mine state features in vibration signal data. The domain discriminator is used to distinguish source domain data from target domain data to assist the feature extractor in mining common features of the source domain data and the target domain data. The new fault detector is used to identify newly occurring fault state signals. The composite fault classifier is used to decouple composite faults.

[0012] Based on a collaborative training strategy, adversarial domain adaptation, open domain adaptation, and pseudo-label learning techniques are integrated to build a multi-task learning framework. The parameters of the feature extractor, domain discriminator, new fault detector, and composite fault classifier are trained and optimized simultaneously to enable the fault diagnosis model to obtain the optimal network parameters.

[0013] A test model is constructed using the trained feature extractor and compound fault classifier and tested on a test set to achieve known fault diagnosis, new fault detection, and compound fault decoupling tasks.

[0014] Furthermore, obtaining a training set and a test set according to the vibration signal data includes:

[0015] Constructing source domain training and test datasets: Collect vibration signal data from mechanical equipment under a certain working condition, cut samples according to the preset sample point length and overlap rate, and attach corresponding labels to form the source domain training dataset; collect vibration signal data from different time periods under the same working condition, cut samples according to the preset sample point length and overlap rate to form the source domain test dataset;

[0016] Construct the target domain training dataset and the target domain test dataset: collect vibration signal data of mechanical equipment under working conditions different from the source domain, intercept samples according to the preset sample point length and overlap rate to form the target domain training dataset; collect vibration signal data of different time periods under the same working conditions, intercept samples according to the preset sample point length and overlap rate to form the target domain test dataset.

[0017] Furthermore, the source domain training data set includes vibration signal data of healthy state and single fault state; the source domain test data set includes vibration signal data of healthy state, single fault state, new fault state and compound fault state;

[0018] The target domain training dataset includes vibration signal data in healthy state, single fault state and new fault state; the target domain test dataset includes vibration signal data in healthy state, single fault state, new fault state and compound fault state.

[0019] Furthermore, the classifier of the new fault detector comprises K+1 neurons, wherein the outputs of 1 to K neurons [p1, p2, ..., p K ] represents the probability of K source domains and target domains sharing categories, and the output p of the K+1th neuron K+1 Represents the probability of a new fault class.

[0020] Furthermore, adversarial domain adaptation technology connects the domain discriminator and feature extractor through a gradient reversal layer, thereby guiding the feature extractor to learn domain-invariant features between different tasks;

[0021] The loss function of adversarial domain adaptation is expressed as:

[0022]

[0023] in, Representative source domain sample x S and target domain samples x T , represents the domain label of the input sample, θ FE represents the trainable parameters in the feature extractor, θ DD represents the trainable parameters in the domain discriminator, N S Represents the number of samples in the source domain, N T represents the number of samples in the target domain, and d represents the domain label; The domain label of the i-th sample is c, d i,c Represents the probability that the i-th sample is predicted to be label c;

[0024] The training process of adversarial domain adaptation is expressed as follows:

[0025]

[0026]

[0027] in, and Represents θ FE and θ DD Optimized parameters.

[0028] Furthermore, the open domain adaptation technology uses labeled source domain samples to train the feature extractor and the new fault detector in a supervised learning manner. During the training process, cross entropy is used as the loss function, and the loss function is expressed as:

[0029]

[0030] Among them, y S represents the source domain sample label, θ EFD represents the trainable parameters in the new fault detector, C=K+1 represents the number of neurons in the classifier of the new fault detector, and p S Represents the pseudo label of the source domain sample;

[0031] In order to achieve the purpose of new fault detection, the output probability of the K+1th neuron of the new fault detector classifier is Set the discrimination boundary t. When the output probability is less than t, the corresponding sample is diagnosed as a known fault state. When the output probability is greater than t, the corresponding sample is diagnosed as a new fault state. The loss function used to implement new fault detection is the binary cross entropy loss function, which is expressed as:

[0032]

[0033] Among them, p T Represents the pseudo label of the target domain sample;

[0034] The training process of open domain adaptation is expressed as follows:

[0035]

[0036]

[0037] in, represents θ EFD The optimized parameters, x T represents the target domain sample.

[0038] Furthermore, the pseudo-label learning technique uses the pseudo-labels generated by the new fault detector to train the composite fault classifier in a semi-supervised learning manner; the predicted probability p of the new fault detector for the i-th unlabeled target sample is i Can be converted into pseudo labels This enables the composite fault classifier to be trained in a supervised manner; the pseudo-label generation process is expressed as follows:

[0039]

[0040] Based on the pseudo labels of the target domain samples, the composite fault classifier can be optimized by minimizing the marginal loss function, which is as follows:

[0041]

[0042] in, is the pseudo label of the target domain sample, λ is the penalty weight, m + and m - Represent the upper and lower boundaries of the predicted probability, v T is the predicted probability of the target domain sample;

[0043] In order to maximize the use of labeled data, the composite fault classifier is optimized based on labeled source domain samples using a marginal loss function. The optimization process is as follows:

[0044]

[0045] Among them, v S is the predicted probability of the source domain sample.

[0046] Furthermore, the overall loss function of the multi-task learning framework is expressed as:

[0047]

[0048] Among them, β, β2, β3, β4, and β5 represent the weight coefficients of the corresponding loss function;

[0049] The importance of each task in the multi-task learning framework is different, so a collaborative optimization strategy based on a dynamic weight algorithm is adopted to simultaneously optimize the trainable parameters of each task; the formula of the dynamic weight algorithm is as follows:

[0050]

[0051] Among them, Epoch represents the number of model optimization iterations, and Num_epoches represents the iteration value set before model training.

[0052] Furthermore, the fault diagnosis model integrates adversarial domain adaptation technology, open domain adaptation technology, and pseudo-label learning technology. The update process of its optimized parameters in each training step is as follows:

[0053]

[0054]

[0055]

[0056]

[0057] Among them, δ represents the learning rate of model update, α represents the penalty factor of the reverse gradient layer, and θ CFC represents the trainable parameters of the composite fault classifier, represents θ CFC Optimized parameters.

[0058] Another technical solution adopted in the present invention is:

[0059] A collaborative diagnosis device for new faults and composite faults, comprising:

[0060] at least one processor;

[0061] at least one memory for storing at least one program;

[0062] When the at least one program is executed by the at least one processor, the at least one processor implements the method described above.

[0063] Another technical solution adopted in the present invention is:

[0064] A computer-readable storage medium stores a program executable by a processor, wherein the program executable by the processor is used to perform the method described above when executed by the processor.

[0065] The beneficial effects of the present invention are as follows: the present invention provides a collaborative diagnosis framework for new and compound faults of mechanical equipment based on an open set capsule network, which can simultaneously realize known fault diagnosis, new fault detection and compound fault decoupling, and effectively reduce the probability of misdiagnosis and missed diagnosis when encountering new and compound faults in actual industrial scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following introduction is made to the drawings of the embodiments of the present invention or the related technical solutions in the prior art. It should be understood that the drawings introduced below are only for the convenience of clearly describing some embodiments of the technical solutions of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative work.

[0067] Figure 1 This is a flow chart of a collaborative diagnosis method for new and complex faults of mechanical equipment based on an open set capsule network in an embodiment of the present invention;

[0068] Figure 2 2 is a schematic diagram of the structure of an open capsule network according to an embodiment of the present invention;

[0069] Figure 3 Schematic diagram of an automobile transmission test bench according to an embodiment of the present invention;

[0070] Figure 4 This is a structural diagram of an automobile transmission in an embodiment of the present invention;

[0071] Figure 5 It is a structural diagram of the model proposed in the embodiment of the present invention;

[0072] Figure 6 Schematic diagram of the confusion matrix of the diagnosis results of scenario A→D in an embodiment of the present invention;

[0073] Figure 7 is a graph of the output probability values ​​of the composite fault classifier for scenario A→D in an embodiment of the present invention;

[0074] Figure 8 Schematic diagram of the confusion matrix of the diagnosis results of scenario A→D in an embodiment of the present invention;

[0075] Figure 9 Schematic diagram of the diagnostic results of the model and comparison method proposed in the embodiment of the present invention;

[0076] Figure 10 Schematic diagram of the confusion matrix of the diagnosis results of the method proposed in the embodiment of the present invention and the comparative method in the collaborative fault diagnosis task. DETAILED DESCRIPTION

[0077] The embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention and are not to be construed as limiting the present invention. The step numbers in the following embodiments are provided for ease of explanation only and do not limit the order of the steps. The order of execution of the steps in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0078] In the description of the present invention, it should be understood that descriptions involving orientations, such as up, down, front, back, left, right, etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, they cannot be understood as limitations on the present invention.

[0079] In the description of the present invention, "several" means one or more, "many" means more than two, "greater than," "less than," and "exceed" are understood to exclude the number itself, while "above," "below," and "within" are understood to include the number itself. The use of "first" and "second" in the description is solely for the purpose of distinguishing technical features and should not be construed as indicating or implying relative importance, implicitly specifying the number of the indicated technical features, or implicitly specifying the order of the indicated technical features.

[0080] Furthermore, in the description of this invention, unless otherwise specified, "plurality" refers to two or more. "And / or" describes the relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can mean: A exists alone, A and B exist simultaneously, or B exists alone. The character " / " generally indicates that the associated objects are in an "or" relationship.

[0081] In the description of the present invention, unless otherwise clearly defined, terms such as setting, installing, and connecting should be understood in a broad sense, and technicians in the relevant technical field can reasonably determine the specific meanings of the above terms in the present invention based on the specific content of the technical solution.

[0082] like Figure 1 As shown, this embodiment provides a collaborative diagnosis method for new faults and composite faults of mechanical equipment based on an open set capsule network. First, a training model is constructed by combining a feature extractor, a domain discriminator, a new fault detector, and a composite fault classifier. Then, a collaborative training strategy is used to optimize the parameters of the training model to the optimal state. Finally, a diagnostic model is constructed using a fully trained feature extractor and a composite fault classifier, and the tasks of known fault diagnosis, new fault detection, and composite fault decoupling are simultaneously achieved. Through the method of this embodiment, the problems of misdiagnosis and missed diagnosis that occur in the intelligent fault diagnosis model when encountering new faults or composite faults of mechanical equipment can be effectively avoided, thereby improving the ability of the intelligent fault diagnosis model to solve practical engineering problems. The structural diagram of the open set capsule network is shown in FIG. Figure 2 As shown, the method includes the steps of:

[0083] Step 1: Construct training and test datasets. Use the accelerometer installed on the mechanical equipment to collect the vibration signal data of the mechanical equipment.

[0084] Construction of source domain training dataset and source domain test dataset: Collect vibration signal data of mechanical equipment under a certain working condition, cut samples according to a certain sample point length and overlap rate, and attach corresponding labels to form the source domain training dataset; collect vibration signal data of different time periods under the same working condition, cut samples according to a certain sample point length and overlap rate to form the source domain test dataset.

[0085] Construction of the target domain training dataset and target domain test dataset: Collect vibration signal data of mechanical equipment under working conditions different from the source domain, intercept samples according to a certain sample point length and overlap rate to form the target domain training dataset; collect vibration signal data of different time periods under the same working conditions, intercept samples according to a certain sample point length and overlap rate to form the target domain test dataset.

[0086] Step 2: Build a training model (i.e., fault diagnosis module). This model consists of a feature extractor, a domain discriminator, a new fault detector, and a composite fault classifier. The feature extractor mines state features in vibration signal data; the domain discriminator distinguishes source and target domain data, helping the feature extractor mine shared features between the source and target domains; the new fault detector identifies newly emerging fault state signals; and the composite fault classifier decouples composite faults.

[0087] Step 3: Optimize model parameters. Based on a collaborative training strategy, we integrate adversarial domain adaptation, open domain adaptation, and pseudo-label learning techniques to build a multi-task learning framework. We simultaneously train and optimize the parameters of the feature extractor, domain discriminator, novel fault detector, and composite fault classifier, ensuring that the fault diagnosis model obtains optimal network parameters.

[0088] Step 4: Collaborative diagnosis of known faults, new faults, and compound faults. A test model is constructed using a fully trained feature extractor and compound fault classifier. Test samples are input into the feature extractor, and the compound fault classifier outputs the diagnosis results, simultaneously achieving known fault diagnosis, new fault detection, and compound fault decoupling tasks.

[0089] As an optional implementation, in step 1, the source domain training dataset includes vibration signal data in healthy and single fault states; the source domain test dataset includes vibration signal data in healthy, single fault, new fault, and compound fault states. The target domain training dataset includes vibration signal data in healthy, single fault, and new fault states; and the target domain test dataset includes vibration signal data in healthy, single fault, new fault, and compound fault states.

[0090] As an optional implementation, in step 2, the classifier of the new fault detector includes K+1 neurons, where the outputs of 1 to K neurons [p1, p2, L, p K ] represents the probability of K source domains and target domains sharing categories, and the output p of the K+1th neuron K+1 Represents the probability of a new fault class.

[0091] As an optional implementation, in step 3, adversarial domain adaptation connects the domain discriminator and feature extractor through a gradient reversal layer, thereby guiding the feature extractor to learn domain-invariant features between different tasks. The loss function of adversarial domain adaptation can be expressed as:

[0092]

[0093] in, Representative source domain sample x S and target domain samples x T , represents the domain label of the input sample, θ FE represents the trainable parameters in the feature extractor, θ DD represents the trainable parameters in the domain discriminator, N S Represents the number of samples in the source domain, N T represents the number of samples in the target domain, and d represents the domain label.

[0094] The training process of adversarial domain adaptation can be expressed as:

[0095]

[0096]

[0097] in, and Represents θ FE and θ DD Optimized parameters.

[0098] As an optional implementation, in step 3, the open domain adaptation technology uses labeled source domain samples to train the feature extractor and the new fault detector in a supervised learning manner. During the training process, the cross entropy is used as the loss function, which can be expressed as:

[0099]

[0100] Among them, y S represents the source domain sample label, θ EFD represents the trainable parameters in the new fault detector, C=K+1 represents the number of neurons in the classifier of the new fault detector, and p S Represents the pseudo label of the source domain sample.

[0101] In order to achieve the purpose of new fault detection, the output probability of the K+1th neuron of the new fault detector classifier is Set the judgment boundary t (0.5≤t≤1). When the output probability is less than t, the corresponding sample is diagnosed as a known fault state. When the output probability is greater than t, the corresponding sample is diagnosed as a new fault state. The loss function used to implement new fault detection is the binary cross entropy loss function, which can be expressed as:

[0102]

[0103] Among them, p T Represents the pseudo label of the target domain sample.

[0104] The training process of open domain adaptation can be expressed as:

[0105]

[0106]

[0107] in, represents θ EFD The optimized parameters, x T represents the target domain sample.

[0108] As an optional implementation, in step 3, the pseudo-label learning technique uses the pseudo-labels generated by the new fault detector to train the composite fault classifier in a semi-supervised learning manner. The predicted probability p of the new fault detector for the i-th unlabeled target sample is i Can be converted into pseudo labels This enables the composite fault classifier to be trained in a supervised manner. The pseudo-label generation process can be expressed as follows:

[0109]

[0110] Based on the pseudo labels of the target domain samples, the composite fault classifier can be optimized by minimizing the marginal loss function, which is as follows:

[0111]

[0112] in, is the pseudo label of the target domain sample, λ is the penalty weight, m + and m - Represent the upper and lower boundaries of the predicted probability, v T is the predicted probability of the target domain sample.

[0113] In order to maximize the use of labeled data, the composite fault classifier can also be optimized using the marginal loss function based on labeled source domain samples. The optimization process is as follows:

[0114]

[0115] Among them, v S is the predicted probability of the source domain sample.

[0116] As an optional implementation, in step 3, the overall loss function of the multi-task learning framework can be expressed as:

[0117]

[0118] Among them, β * Represents the weight coefficient of the corresponding loss function.

[0119] The importance of each task in the multi-task learning framework is different, so a collaborative optimization strategy based on the dynamic weight algorithm is adopted to optimize the trainable parameters of each task simultaneously. The formula of the dynamic weight algorithm is as follows:

[0120]

[0121] Among them, Epoch represents the number of model optimization iterations, and Num_epoches represents the iteration value set before model training.

[0122] The proposed model integrates adversarial domain adaptation, open domain adaptation, and pseudo-label learning techniques. The update process of its optimized parameters in each training step is as follows:

[0123]

[0124]

[0125]

[0126]

[0127] Among them, δ represents the learning rate of model update, α represents the penalty factor of the reverse gradient layer, and θ CFC represents the trainable parameters of the composite fault classifier, represents θ CFC Optimized parameters.

[0128] As an optional implementation, in step 4, the key steps of the proposed collaborative diagnosis framework for known faults, new faults, and compound faults are shown in Table 1.

[0129] Table 1. Key steps of the collaborative fault diagnosis framework

[0130]

[0131]

[0132] Experimental case:

[0133] Key components of automotive transmissions, including bearings and gears, are prone to new and combined faults during long-term service. Data collected from a five-speed transmission of a certain vehicle model is used to evaluate and verify the effectiveness of the proposed method.

[0134] 1. Experimental platform and dataset

[0135] The structure of the test bench and automobile transmission is as follows Figure 3 and Figure 4 As shown in Figure 2, raw vibration signals were collected from four different operating conditions using a BBM-PAK data logger at a sampling frequency of 24 kHz. Typical health states collected included: normal state (HM), minor gear tooth breakage fault (GF), bearing inner race fault (BF), and composite fault (CF, a combination of GF and BF). The collected health state data was used to simulate new fault detection and composite fault decoupling scenarios under different speeds and loads. The automotive transmission dataset is shown in Table 2.

[0136] In the experiment, three representative scenarios covering different fault conditions were designed to simulate the collaborative fault diagnosis task and cross-validate the proposed model. The fault conditions in the experiment included: 1) different loads, 2) different speeds, and 3) different speeds and loads. For simplicity, condition A (1000 rpm / 50 Nm) was used as the source domain. In the source domain, the model was trained in a supervised manner using labeled training samples of HM and GF. One of the remaining three conditions was used as the target domain. In the target domain, the model was trained using unlabeled training samples of HM, GF, and BF. BF is a fault state that does not appear in the source domain but appears in the target domain, and is therefore considered a new fault. For the composite fault decoupling task, only healthy and single fault samples from the source and target domains were used to train the model. In other words, during the model testing phase, the composite fault type is unknown to the model. The cross-validation settings for collaborative fault diagnosis are shown in Table 3.

[0137] Table 2 Automobile transmission dataset

[0138]

[0139]

[0140] Table 3 Cross-validation settings for collaborative fault diagnosis

[0141]

[0142] 2. Structure and parameters of the proposed model

[0143] The structure of the proposed model is as follows Figure 5As shown in Table 3, based on the experimental settings, the value of K (the number of labeled fault types in the source domain) in the model architecture was set to 2, so C = K + 1 = 3. The model was built using the PyTorch tensor computing framework and GPU-accelerated using an NVIDIA TITAN Xp graphics card. The model parameters were optimized using the Nadam optimizer. The model hyperparameters were set as follows: learning rate 5e-4, batch size 64, and epochs 50.

[0144] 3. Collaborative fault diagnosis results of the proposed model

[0145] Three cross-validation experiments are proposed to simulate the three representative scenarios mentioned above to evaluate the diagnostic performance of the proposed model. The evaluation metric for collaborative fault diagnosis performance is diagnostic accuracy, as shown in Table 4. The diagnostic accuracy of each task (i.e., known fault classification, new fault detection, and compound fault decoupling) is calculated based on source and target domain samples during the model testing phase.

[0146] Table 4 Collaborative fault diagnosis results of the proposed model

[0147]

[0148] The results in Table 4 demonstrate that the proposed model is not only capable of identifying known labeled categories in the source domain, but also capable of detecting novel faults occurring in both the source and target domains, achieving average diagnostic accuracies of 98.60% and 100.00%, respectively. Furthermore, even when compound fault sample data was not present during the model's training phase, the proposed model was able to correctly diagnose compound faults (CF) coupled with known faults (GF) and novel faults (BF), achieving a diagnostic accuracy of 99.35%. Furthermore, the average diagnostic accuracy for all collaborative fault diagnosis scenarios reached 99.14%, demonstrating that the proposed model is capable of collaboratively and effectively performing multi-fault diagnosis tasks within the same diagnostic model. These results demonstrate that the proposed method provides an effective solution to the challenging problem of misdiagnosis or missed diagnosis caused by novel and compound faults.

[0149] Specifically, taking the cross-validation scenario A→D with different speeds and loads as an example, Figure 6The confusion matrix of the collaborative diagnosis results is shown. Among them, the predicted label "OC (other combinations)" represents other possible combinations of multiple output labels in addition to the true label. From the confusion matrix, it can be observed that for compound faults, 1.4% of the samples in the source domain are diagnosed as BF, and 0.28% of the samples in the target domain are diagnosed as OC; for new faults, the diagnosis results of the proposed model in both the source domain and the target domain can reach 100%. Although the diagnostic accuracy of compound faults does not reach 100%, overall, the proposed model can still effectively and collaboratively complete the tasks of known fault classification, new fault detection, and compound fault decoupling diagnosis.

[0150] In addition, in order to help readers understand the proposed method more clearly, the output probability and corresponding output label of the composite fault classifier (CFC) are plotted, as shown in Figure 7 and Figure 8 As shown in the figure. For each test sample, the red line (denoted by TS) in the figure represents the confidence threshold, which is used to determine whether to output the corresponding label. Generally, the output probability of each class neuron represents the probability that the input sample belongs to the corresponding class. For simplicity, TS is set to 0.5, that is, when the predicted probability of the class neuron is greater than 0.5, the corresponding label is output.

[0151] Figure 8 Among them, samples 1-357, 358-714, 715-1071 and 1072-1428 belong to the test samples of HM, GF, BF and CF respectively. Figure 7 and Figure 8 The experimental data shows that the proposed model can accurately classify most test samples into the correct category under different speed and load scenarios. Specifically, the proposed model can effectively decouple the composite fault type CF (coupled by BF and GF), which does not appear in the training set, into two single faults, BF and GF.

[0152] 4. Comparison method and fault diagnosis results

[0153] In order to further demonstrate the effectiveness and superiority of the proposed method, several advanced methods in the field are used for fair comparison and discussion. (1) WDCNN: A deep learning-based intelligent fault diagnosis method trained in a supervised manner. (2) DANN: A deep migration method that uses domain adaptation technology to improve the generalization performance of the diagnosis model. (3) DATLN: A fault diagnosis method based on open domain adaptation is applied to detect new faults of gears and bearings. This method constructs a classifier that can learn decision boundaries to detect new faults from unlabeled training samples. (4) DDCNN: An effective intelligent fault diagnosis method is used to decouple the composite faults of mechanical equipment. Considering the assumptions of DDCNN, two training settings are used for comparison, denoted as DDCNN and DDCNN+EF.

[0154] The experimental setup, including training and test sets, is shown in Table 5. Deep learning-based methods (WDCNN, DDCNN, and DDCNN+EF) are trained using labeled samples from the source domain; deep transfer learning-based methods (DANN, DATLN, and the proposed method) are trained using labeled samples from the source domain and unlabeled samples from the target domain. Considering the data dependency of the composite fault decoupling task, DDCNN is trained under two conditions: one lacking new fault BF samples in the training set, and the other containing new fault BF samples. The corresponding methods are denoted as DDCNN and DDCNN+EF, respectively. It should be noted that, in the problem assumptions, labeled BF samples are unavailable.

[0155] Table 5 Experimental settings of the proposed model and the compared methods

[0156]

[0157] The diagnostic results of the proposed model and the comparison method are shown in Table 6 and Figure 9 The diagnostic results are shown in Figure 2. The numerical values ​​of the diagnostic results are the average test accuracy of three cross-validation scenarios in the source and target domains. Overall, the proposed model achieved the best diagnostic accuracy of 99.14% when performing the three collaborative diagnosis tasks, while the average accuracy of the other five baseline methods were 49.25%, 49.91%, 73.38%, 48.43%, and 98.29%, respectively.

[0158] Table 6 Diagnostic results of the proposed model and the comparison method

[0159]

[0160] Specifically, due to their lack of novel fault detection and compound fault decoupling capabilities, WDCNN and DANN can only identify labeled known class samples during model training. By introducing open domain adaptation techniques, DATLN is able to effectively detect novel faults and achieve a high diagnostic accuracy of 99.11%. However, this approach is unable to identify and decouple compound faults when compound fault samples are unavailable. Furthermore, while DDCNN+EF achieves higher diagnostic accuracy than the proposed method for novel fault detection and compound fault decoupling, the superior performance of DDCNN+EF relies on the completeness of labeled BF samples, which is impractical when performing novel fault detection. Analysis of the diagnostic results of DDCNN and DDCNN+EF reveals that if labeled novel fault samples are unavailable for supervised model training, the diagnostic performance of DDCNN+EF will significantly degrade, and it may even be unable to perform both novel fault detection and compound fault decoupling simultaneously.

[0161] Similarly, taking A→D as an example, Figure 10 The confusion matrix of the proposed method and the comparison methods in the collaborative fault diagnosis task is shown. The values ​​in the confusion matrix represent the predicted probabilities of the model output. It can be seen that, through supervised learning, all six methods can effectively diagnose the HM and GF categories. However, three of the methods (WDCNN, DANN, and DDCNN) completely fail in the new fault detection and composite fault decoupling tasks, where the new fault BF is misdiagnosed as HM and the composite fault CF is missed as GF or OC. In the absence of labeled new fault samples, DATLN can also effectively detect new faults, but it cannot achieve composite fault decoupling, as DATLN misdiagnoses the CF category as GF (28.01%) and BF (71.99%). In contrast, the proposed model achieves excellent diagnostic accuracy in all diagnosis tasks in the absence of labeled new fault samples, and is competitive even when compared to the experimental results of DDCNN+EF trained with labeled new fault samples.

[0162] Through experimental comparison and discussion, the following conclusions are drawn: By combining open domain adaptation technology and pseudo-label learning technology, the proposed collaborative fault diagnosis model can effectively perform known fault classification, new fault detection and compound fault decoupling tasks simultaneously. By imitating human multi-task learning ability, the intelligence and practicality of the intelligent fault diagnosis method are further improved.

[0163] In summary, the method of the present invention has the following advantages and beneficial effects compared to the prior art:

[0164] (1) The present invention provides a collaborative diagnosis framework for new and compound faults of mechanical equipment based on an open set capsule network, which can simultaneously realize known fault diagnosis, new fault detection and compound fault decoupling, effectively reducing the probability of misdiagnosis and missed diagnosis when encountering new and compound faults in actual industrial scenarios.

[0165] (2) The present invention provides a collaborative optimization strategy that integrates adversarial domain adaptation, open domain adaptation, and pseudo-label learning technologies to optimize model parameters and give the model the ability to simultaneously detect new faults and decouple complex faults.

[0166] (3) The present invention proposes a new fault and compound fault collaborative diagnosis method by simulating human multi-task learning ability, which further improves the intelligence, accuracy and efficiency of the intelligent fault diagnosis model in actual industrial applications.

[0167] This embodiment also provides a new fault and compound fault collaborative diagnosis device, including:

[0168] at least one processor;

[0169] at least one memory for storing at least one program;

[0170] When the at least one program is executed by the at least one processor, the at least one processor implements the following Figure 1 The method shown.

[0171] A new fault and composite fault collaborative diagnosis device of this embodiment can execute a new fault and composite fault collaborative diagnosis method provided by the method embodiment of the present invention, can execute any combination of implementation steps of the method embodiment, and has the corresponding functions and beneficial effects of the method.

[0172] The present application also discloses a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device performs Figure 1 The method shown.

[0173] This embodiment also provides a storage medium that stores instructions or programs that can execute a new fault and composite fault collaborative diagnosis method provided by the method embodiment of the present invention. When the instructions or programs are run, any combination of implementation steps of the method embodiment can be executed, and the corresponding functions and beneficial effects of the method can be obtained.

[0174] In some optional embodiments, the function / operation mentioned in the block diagram may not occur in the order mentioned in the operation diagram. For example, depending on the function / operation involved, the two boxes shown in succession can actually be executed substantially simultaneously or the boxes can sometimes be executed in reverse order. In addition, the embodiment presented and described in the flow chart of the present invention is provided in an exemplary manner for the purpose of providing a more comprehensive understanding of the technology. The disclosed method is not limited to the operation and logic flow presented herein. Optional embodiments are contemplated in which the order of the various operations is changed and the sub-operations described as a part of a larger operation are performed independently.

[0175] Furthermore, although the present invention is described in the context of functional modules, it should be understood that, unless otherwise indicated, one or more of the functions and / or features described may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in separate physical devices or software modules. It will also be understood that a detailed discussion of the actual implementation of each module is not necessary for understanding the present invention. More specifically, given the properties, functions, and internal relationships of the various functional modules in the devices disclosed herein, the actual implementation of the module will be understood within the ordinary skill of an engineer. Therefore, a person skilled in the art using ordinary skill will be able to implement the present invention set forth in the claims without undue experimentation. It will also be understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of the present invention, which is determined by the full scope of the appended claims and their equivalents.

[0176] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0177] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0178] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0179] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0180] In the above description of this specification, reference to the terms "one embodiment / example," "another embodiment / example," or "certain embodiments / examples" means that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0181] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to the embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the claims and their equivalents.

[0182] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.

Claims

1. A new fault and compound fault collaborative diagnosis method, characterized in that: The following steps are involved: Collect vibration signal data of mechanical equipment and obtain training sets and test sets based on the vibration signal data; A fault diagnosis model is constructed and trained using a training set. The fault diagnosis model includes a feature extractor, a domain discriminator, a new fault detector, and a composite fault classifier. The feature extractor is used to mine state features in vibration signal data. The domain discriminator is used to distinguish source domain data from target domain data to assist the feature extractor in mining common features of the source domain data and the target domain data. The new fault detector is used to identify newly occurring fault state signals. The composite fault classifier is used to decouple composite faults. Based on a collaborative training strategy, adversarial domain adaptation, open domain adaptation, and pseudo-label learning techniques are integrated to build a multi-task learning framework. The parameters of the feature extractor, domain discriminator, new fault detector, and composite fault classifier are trained and optimized simultaneously to enable the fault diagnosis model to obtain the optimal network parameters. Use the trained feature extractor and compound fault classifier to build a test model and test it on the test set to achieve known fault diagnosis, new fault detection and compound fault decoupling tasks; Adversarial domain adaptation technology connects the domain discriminator and feature extractor through a gradient reversal layer, thereby guiding the feature extractor to learn domain-invariant features between different tasks; The loss function of adversarial domain adaptation is expressed as: in, Representative source domain sample x S and target domain samples x T , represents the domain label of the input sample, θ FE represents the trainable parameters in the feature extractor, θ DD represents the trainable parameters in the domain discriminator, N S Represents the number of samples in the source domain, N T represents the number of samples in the target domain, and d represents the domain label; The domain label of the i-th sample is c, d i,c Represents the probability that the i-th sample is predicted to be label c; The training process of adversarial domain adaptation is expressed as follows: in, and Represents θ FE and θ DD Optimized parameters.

2. A new fault and compound fault collaborative diagnosis method according to claim 1, characterized in that: The step of obtaining a training set and a test set based on the vibration signal data includes: Constructing source domain training and test datasets: Collect vibration signal data from mechanical equipment under a certain working condition, cut samples according to the preset sample point length and overlap rate, and attach corresponding labels to form the source domain training dataset; collect vibration signal data from different time periods under the same working condition, cut samples according to the preset sample point length and overlap rate to form the source domain test dataset; Construct the target domain training dataset and the target domain test dataset: collect vibration signal data of mechanical equipment under working conditions different from the source domain, intercept samples according to the preset sample point length and overlap rate to form the target domain training dataset; collect vibration signal data of different time periods under the same working conditions, intercept samples according to the preset sample point length and overlap rate to form the target domain test dataset.

3. A new fault and compound fault collaborative diagnosis method according to claim 2, characterized in that: The source domain training dataset includes vibration signal data in healthy state and single fault state; the source domain test dataset includes vibration signal data in healthy state, single fault state, new fault state and compound fault state; The target domain training dataset includes vibration signal data in healthy state, single fault state and new fault state; the target domain test dataset includes vibration signal data in healthy state, single fault state, new fault state and compound fault state.

4. A new fault and compound fault collaborative diagnosis method according to claim 3, characterized in that: The classifier of the new fault detector includes K+1 neurons, wherein the outputs of 1 to K neurons [p1, p2, ..., p K ] represents the probability of K source domains and target domains sharing categories, and the output p of the K+1th neuron K+1 Represents the probability of a new fault class.

5. A new fault and compound fault collaborative diagnosis method according to claim 4, characterized in that: The open domain adaptation technology uses labeled source domain samples to train the feature extractor and new fault detector in a supervised learning manner. During the training process, cross entropy is used as the loss function. The expression of the loss function is: Among them, y S represents the source domain sample label, θ EFD represents the trainable parameters in the new fault detector, C=K+1 represents the number of neurons in the classifier of the new fault detector, and p S Represents the pseudo label of the source domain sample; In order to achieve the purpose of new fault detection, the output probability of the K+1th neuron of the new fault detector classifier is Set the discrimination boundary t. When the output probability is less than t, the corresponding sample is diagnosed as a known fault state. When the output probability is greater than t, the corresponding sample is diagnosed as a new fault state. The loss function used to implement new fault detection is the binary cross entropy loss function, which is expressed as: Among them, p T Represents the pseudo label of the target domain sample; The training process of open domain adaptation is expressed as follows: in, represents θ EFD The optimized parameters, x T represents the target domain sample.

6. A new fault and compound fault collaborative diagnosis method according to claim 5, characterized in that: The pseudo-label learning technique uses the pseudo-labels generated by the new fault detector to train the composite fault classifier in a semi-supervised learning manner; the new fault detector predicts the probability p of the i-th unlabeled target sample i Can be converted into pseudo labels This enables the composite fault classifier to be trained in a supervised manner; the pseudo-label generation process is expressed as follows: Based on the pseudo labels of the target domain samples, the composite fault classifier can be optimized by minimizing the marginal loss function, which is as follows: in, is the pseudo label of the target domain sample, λ is the penalty weight, m + and m - Represent the upper and lower boundaries of the predicted probability, v T is the predicted probability of the target domain sample; In order to maximize the use of labeled data, the composite fault classifier is optimized using the marginal loss function based on labeled source domain samples. The optimization process is as follows: Among them, v S is the predicted probability of the source domain sample.

7. A new fault and compound fault collaborative diagnosis method according to claim 6, characterized in that: The overall loss function of the multi-task learning framework is expressed as: Among them, β1, β2, β3, β4, and β5 represent the weight coefficients of the corresponding loss function; The importance of each task in the multi-task learning framework is different, so a collaborative optimization strategy based on a dynamic weight algorithm is adopted to simultaneously optimize the trainable parameters of each task; the formula of the dynamic weight algorithm is as follows: Among them, Epoch represents the number of model optimization iterations, and Num_epoches represents the iteration value set before model training.

8. A new fault and compound fault collaborative diagnosis device, characterized in that: include: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method according to any one of claims 1 to 7.

9. A computer-readable storage medium storing a program executable by a processor, characterized in that: The processor-executable program is used to perform the method according to any one of claims 1 to 7 when executed by the processor.

Images