Hoisting mechanism bearing fault diagnosis method based on causal feature learning

Abstract

The invention relates to the technical field of hoisting mechanism bearing fault diagnosis, in particular to a hoisting mechanism bearing fault diagnosis method based on causal feature learning. Comprising the steps of obtaining bearing source domain data; training a deep learning fault diagnosis model by using the bearing source domain data; obtaining bearing target domain data; and inputting bearing target domain data into the trained deep learning fault diagnosis model, and outputting a bearing fault diagnosis result. According to the hoisting mechanism bearing fault diagnosis method based on the field generalization technology, the problem that an original model cannot perform high-quality and high-efficiency fault diagnosis on new data due to working condition changes in a real environment can be solved, a set of new fault diagnosis method is provided, the data acquisition cost can be remarkably reduced, and the fault diagnosis efficiency is improved. And the safety and the reliability of bearing operation of the hoisting mechanism are greatly improved.

Description

Technical Field

[0001] The invention relates to the technical field of hoisting mechanism bearing fault diagnosis, and in particular to a hoisting mechanism bearing fault diagnosis method based on causal feature learning. Background Art

[0002] Lifting mechanisms are widely used in various types of lifting equipment and material handling systems, and are used to perform different tasks of vertically lifting and lowering heavy objects. In various factory workshops, construction sites, port terminals and other environments, they are responsible for carrying various heavy materials or finished products, and installing and shifting large structural parts. These tasks usually require the lifting mechanism to safely and reliably handle loads of different weights and sizes to ensure that the materials can be placed accurately at the designated location. In this process, the lifting mechanism bearing, as a key component of the lifting mechanism, plays an indispensable role. It can support and guide moving parts, reduce friction, and ensure the smooth operation of the transmission system. Especially under high load conditions, the quality and performance of the bearing are directly related to the stability and work efficiency of the entire system. A healthy bearing can not only withstand pressure from all directions, but also adapt to different temperature changes and other harsh working conditions, ensuring that the equipment can operate stably for a long time. However, during the long-term high-load use of the bearing, especially when the employee is not operating in a standardized manner, the lifting mechanism jog phenomenon often occurs, and the bearing is very prone to fatigue or early weak failures of varying degrees. Limited by the installation position of the bearing and the overall shape of the lifting mechanism, it is usually difficult to directly check the health of the bearing. Existing health management methods usually adopt the method of regular replacement, that is, the equipment is replaced at certain working intervals. Unfortunately, under different loads and usage habits, even for the same batch of bearings produced by the same manufacturer, the remaining service life of the hoisting mechanism bearings is still quite different. If the bearing service life is close to the threshold and is not replaced, it may cause increased mechanical vibration, affect the operating accuracy, and even damage the entire equipment, and even endanger the lives of on-site personnel. If the bearing is far from reaching the remaining service life threshold, that is, replacing the bearing will cause serious waste of resources and increase production costs. Therefore, effective status monitoring and intelligent diagnosis of the hoisting mechanism, especially the bearings therein, is crucial to ensure operational safety and economic benefits.

[0003] With the rapid development of computer, sensor, artificial intelligence, embedded, communication and other technologies, fault diagnosis technology based on deep learning has developed rapidly. However, for hoisting mechanism bearings, there are still many problems to be solved in its intelligent fault diagnosis technology. First, there are few fault samples of hoisting mechanism bearings, and due to changes in working conditions, there is a large domain drift between training samples and test samples. For example, data collected under heavy load and low speed conditions and neural networks trained based on this data may not be able to accurately judge the actual health status of light load and high speed samples. In response to this problem, some scholars have proposed different diagnostic methods, such as using domain adaptation technology to extract the common features of source domain data and target domain data by training neural networks, and training the common features to make them discriminative. Existing literature shows that this method can indeed alleviate the problem of reduced fault diagnosis accuracy caused by domain drift. However, it should be noted that this method can only generalize diagnostic knowledge from source domain data to target domains that can participate in training. Once the working conditions of the target domain change again, the fault diagnosis accuracy based on the domain adaptation method will decrease to varying degrees, and there is still a certain lag in retraining the network for newly collected data. Secondly, the domain adaptation-based method still needs to collect target domain data for training, and cannot perform online diagnosis on the data of the actual operation process, which makes the intelligent diagnosis still have a certain lag, and the equipment often "runs with problems".

[0004] The prior art has the following defects:

[0005] 1. The existing technology cannot effectively utilize the huge and complex data, and requires the subjective experience of the operator to judge the health status of the lifting mechanism bearing, and the diagnostic efficiency and success rate are low.

[0006] 2. Fault diagnosis based on deep learning methods cannot meet the reliability requirements in the case of data drift.

[0007] 3. Existing technologies rely on a large number of labeled high-quality data sets, but the cost of data collection is too high, and the collected data usually does not contain a large number of labels, resulting in low diagnostic accuracy of existing methods.

[0008] 4. Fault diagnosis based on domain adaptive methods cannot cope with the real-time data generated during industrial operation, and once the working conditions change, the diagnostic accuracy will still decrease to varying degrees.

[0009] 5. Traditional methods based on domain generalization only rely on statistical information of operating data and its corresponding health status for modeling, and fail to extract causal information that determines the health status of specific sample data, resulting in diagnostic accuracy that cannot meet requirements. Summary of the invention

[0010] The present invention discloses a method for diagnosing bearing faults of a hoisting mechanism based on causal feature learning, and the specific method is as follows:

[0011] Obtain bearing source domain data;

[0012] Use bearing source domain data to train a deep learning fault diagnosis model;

[0013] Acquire bearing target domain data;

[0014] The bearing target domain data is input into the trained deep learning fault diagnosis model, and the bearing fault diagnosis results are output.

[0015] Furthermore, when training the deep learning fault diagnosis model with bearing source domain data, the customized total loss function L Total , specifically including:

[0016] Evaluation item L of diagnostic accuracy of deep learning fault diagnosis model ce ;

[0017] The source domain data similarity evaluation item L before and after the introduction of interference signals sim ;

[0018] Before and after the introduction of the interference signal, the same source domain data is processed by the deep learning fault diagnosis model, and the similarity evaluation item L of the output feature dimension is dep ;

[0019] The similarity evaluation item L of the bearing source domain data and the bearing target domain data after extracting features through the deep learning fault diagnosis model ang .

[0020] Furthermore, the total loss function L is customized ce , the specific formula is as follows:

[0021]

[0022] In the formula, α, β, γ are all hyperparameters, and N is the number of samples.

[0023] Furthermore, the diagnostic accuracy evaluation item L of the deep learning fault diagnosis model ce , the Softmax classification layer adopts the cross entropy loss function, the specific formula is as follows:

[0024]

[0025] in, is the probability distribution calculated from the original output of the model by the Softmax function, y c is the true label, and R is the number of sample categories.

[0026] Furthermore, the source domain data similarity evaluation item Lsim The specific acquisition method is as follows:

[0027] Fourier transform of bearing source domain data;

[0028] Performing perturbation operation on the bearing source domain data after Fourier transformation;

[0029] Perform inverse Fourier transform on the bearing source domain data after the disturbance operation;

[0030] The Pearson correlation coefficient is used to calculate the similarity between the original bearing source domain data and the bearing source domain data after Fourier inverse transformation, which is used as the source domain data similarity evaluation item L sim .

[0031] Furthermore, the similarity evaluation item L of the output feature dimension is dep The specific acquisition method is as follows:

[0032] Fourier transform of bearing source domain data;

[0033] Performing perturbation operation on the bearing source domain data after Fourier transformation;

[0034] Perform inverse Fourier transform on the bearing source domain data after the disturbance operation;

[0035] The Ehrlich correlation coefficient is used to calculate the similarity of the output features of the fully connected layer of the deep learning fault diagnosis model for the same bearing source domain data before and after interference, which is used as the similarity evaluation item L of the output feature dimension. dep .

[0036] Furthermore, the similarity evaluation item L is used for the bearing source domain data and the bearing target domain data after the features are extracted by the deep learning fault diagnosis model. ang The specific acquisition method is as follows:

[0037] Extract feature vectors of bearing source domain data using deep learning fault diagnosis model;

[0038] Extract feature vectors of bearing target domain data using deep learning fault diagnosis model;

[0039] The angular distance metric is used to calculate the similarity of the two feature vectors, which is used as the similarity evaluation item L after the bearing source domain data and the bearing target domain data are extracted by the deep learning fault diagnosis model. ang .

[0040] Furthermore, the bearing source domain data is obtained through a bearing fault test and is labeled data;

[0041] The bearing target domain data is acquired by detecting the actual working site of the lifting mechanism through sensors and is unlabeled data.

[0042] Furthermore, the bearing failure of the lifting mechanism includes: a bearing inner ring failure and a bearing outer ring failure.

[0043] Due to the adoption of the above technical solution, the present invention has the following beneficial effects:

[0044] 1. The hoisting mechanism bearing fault diagnosis method based on domain generalization technology proposed in the present invention can solve the problem that the original model cannot perform high-quality and efficient fault diagnosis on new data due to changes in working conditions under real-world conditions. A new fault diagnosis method is proposed, which can significantly reduce the data collection cost and greatly improve the safety and reliability of the hoisting mechanism bearing operation.

[0045] 2. The domain generalization fault diagnosis method based on causal feature consistency extraction can extract the most essential features that affect the health status of the lifting mechanism bearing, rather than statistical features. It no longer requires that the domain invariant features be absolutely invariant features, but allows the features to have certain fluctuations in value, but the direction is as similar as possible. It is more in line with the actual operating logic of the lifting mechanism bearing and has high diagnostic accuracy.

[0046] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The accompanying drawings of the present invention are as follows.

[0048] Figure 1 It is a schematic diagram of the overall process of the present invention. DETAILED DESCRIPTION

[0049] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0050] A bearing fault diagnosis method for hoisting mechanism based on causal feature learning, such as Figure 1 As shown, the specific steps are as follows:

[0051] S1. Obtain bearing source domain data.

[0052] In this embodiment, the source domain data set is derived from laboratory bearing operation data collected and recorded by an acceleration sensor, which is labeled data. According to the type of common bearing failures, it is divided into inner ring failure data and outer ring failure data, and the operation data under normal conditions is collected to form the source domain data.

[0053] S2. Train the deep learning fault diagnosis model with bearing source domain data.

[0054] In step S2, the main parameters of the deep learning fault diagnosis model are shown in the following table:

[0055]

[0056] When training the deep learning fault diagnosis model with bearing source domain data, the custom total loss function L Total , specifically including:

[0057] S21, deep learning fault diagnosis model diagnostic accuracy evaluation item L ce .

[0058] Evaluation item L of diagnostic accuracy of deep learning fault diagnosis model ce The cross entropy loss function is used in the Softmax classification layer to calculate the health status classification loss of the source domain samples, as follows:

[0059]

[0060] in, is the probability distribution calculated from the original output of the model by the Softmax function, y c is the true label, and R is the number of sample categories.

[0061] S22, source domain data similarity evaluation item L before and after the introduction of interference signal sim ;

[0062] In order to extract the causal factors that determine the health status of the sample from the laboratory's operating data, ignore the interference of non-causal information on the model, and retain as much domain-specific information of the working status as possible, so that the features extracted by the network are consistent with the changes in the working status of the lifting mechanism, the causal information extraction module is first used to model the inherent properties of the ideal causal information so that the extracted features have ideal causal properties.

[0063] Since the phase information of the bearing vibration signal after Fourier transformation can better reflect the intrinsic structural mode of the mechanical system, has higher stability and robustness, and can capture more complex nonlinear interaction effects, it is believed that the phase signal can better represent the causal information, while the amplitude signal contains a large amount of low-level statistical information.

[0064] This embodiment relies on suppressing the effect of amplitude signals and highlighting the role of phase signals in fault diagnosis, thereby extracting essential information that characterizes the health status. In order to capture the relative time relationship and coordinated motion mode between the internal components of the hoisting mechanism bearing, reveal the essential characteristics of the operation of the hoisting mechanism bearing, extract more high-level semantic features and suppress the interference of statistical information, the source domain data is transformed using Fourier interference as follows:

[0065]

[0066] Among them, x 0 is the original sample, F represents Fourier transform, A is the amplitude after Fourier transform, and P is the phase signal after Fourier transform.

[0067] After that, the source domain data is perturbed to generate the induced samples. The specific formula is as follows:

[0068]

[0069] where x 0 ' represents a sample of another source domain data, λ represents the degree of disturbance of the two source domain signals, and λ~(-1,1), Represents the amplitude of the newly generated data after Fourier transformation, and the phase is still the same as the previous sample. After that, the newly generated data is inversely transformed by Fourier transformation. The specific formula is as follows:

[0070]

[0071] Among them, F -1 is the inverse Fourier transform.

[0072] Afterwards, the Pearson correlation coefficient is used to calculate the similarity between the new data and the original data, and this is used as the loss function to induce the features extracted by the deep learning fault diagnosis model to retain the phase information to the maximum extent and suppress the amplitude information. The details are as follows:

[0073]

[0074] Among them, PEA(·,·) represents the calculation of Pearson correlation coefficient, N is the dimension of the features extracted by the neural network, Represents C under specific conditions j The original data and the newly generated data are processed in the high-dimensional space of the network using the MaxAbs Scaling method for the eigenvalues ​​of the i-th dimension, and it is expected that the characteristics of the generated data will have the highest similarity with the characteristics of the original data. The purpose of this formula is to enable the network to retain the advanced physical features related to the operating status of the hoisting mechanism bearing to the maximum extent and eliminate non-causal features.

[0075] S23, before and after the introduction of interference signals, the same source domain data is processed by the deep learning fault diagnosis model, and the similarity evaluation item L of the output feature dimension is dep .

[0076] According to the basic properties of ideal causal features, the feature vectors of causal factors should be independent of each other. In order to simulate the characteristics with such properties, the Pearson correlation coefficient method is used to calculate the similarity between different dimensions of the same vector before and after Fourier interference, and this similarity function is used as the loss function. It is expected that the feature similarity between different dimensions of the features extracted by the neural network is the lowest, as follows:

[0077]

[0078] S24, similarity evaluation item L of bearing source domain data and bearing target domain data after extracting features through deep learning fault diagnosis model ang .

[0079] The operating signal of the hoisting mechanism bearing is jointly determined by the bearing model and the operating conditions. In the existing transfer learning method, the use of feature distribution difference measurement to extract domain-invariant features may ignore the individual differences between different fields. This method has a certain performance upper limit. Therefore, this embodiment proposes a new method that no longer emphasizes the absolute consistency in the numerical sense extracted from the personalized field, but uses the angular distance measurement to measure the similarity of the features of the two source fields, as follows:

[0080]

[0081] Among them, ||R A || and ||R B || represents the Euclidean norm of source domain A and source domain B. Its value range is [0, Π]. When its value is 0, it means that the feature directions of the two source domains are completely consistent, and Π means that the directions are completely opposite.

[0082] S25. In order to ensure that all loss items are on the same scale and avoid some loss items dominating the optimization process due to numerical differences, all losses are normalized. First, for the Pearson similarity loss, it is converted to 1-|r|, where r is the absolute value of the Pearson similarity. Secondly, for the diagonal distance loss, it is mapped to [0,1] by dividing by Π.

[0083] S26. The three similarity functions are combined by weighted summation to obtain the causal feature loss function, which is as follows:

[0084]

[0085] Among them, α, β, γ are all hyperparameters, and N is the number of samples.

[0086] S27. The overall optimization goal is that the extracted features have the basic properties of ideal causal features, that is, the causal factors are independent of each other, the causal factors can be separated from the non-causal factors, and the causal factors are consistent with the actual operating mechanism of the lifting mechanism bearing, that is, different fields have both personalized factors and certain common characteristics. Therefore, the overall loss of the network is as follows:

[0087] L Total =L ce +nL cau

[0088] Among them, n is a hyperparameter that determines the degree of generalization of the field. After clarifying the above loss function, that is, the optimization goal, let θ f ,θ c Represent the parameters of the domain shared feature extractor and the health status classifier respectively, then the total loss function can be re-given by the following formula:

[0089]

[0090] The deep learning fault diagnosis model is trained using stochastic gradient descent and updates the network parameters through back propagation. Using the stochastic gradient descent method, the θ f ,θ c The formula for the process is as follows:

[0091]

[0092] Among them, ε is the learning rate.

[0093] S3. Obtain bearing target domain data.

[0094] In this embodiment, the target domain data is the actual operation data of the hoisting mechanism collected on site using an acceleration sensor. The target domain data is automatically collected by the hoisting mechanism dynamic equipment monitoring system. When the dynamic equipment monitoring system collects operation data, it may be disturbed by wind, abnormal operation of on-site operators, abnormal bumpy working environment, etc., causing the sensing equipment to collect noise signals of varying degrees, or being interfered by electromagnetic signals or data missing during data transmission and storage, which affects the performance of the intelligent fault diagnosis model. Therefore, before inputting the data into the model, the bearing vibration signal is first denoised and missing values ​​are filled by data cleaning.

[0095] S4. Input the bearing target domain data into the trained deep learning fault diagnosis model and output the bearing fault diagnosis results.

[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for diagnosing bearing faults of a lifting mechanism based on causal feature learning, characterized in that: The specific method is as follows: Obtain bearing source domain data; Use bearing source domain data to train a deep learning fault diagnosis model; Acquire bearing target domain data; The bearing target domain data is input into the trained deep learning fault diagnosis model, and the bearing fault diagnosis results are output.

2. The method for diagnosing bearing faults of a lifting mechanism based on causal feature learning according to claim 1, characterized in that: When training the deep learning fault diagnosis model with bearing source domain data, the custom total loss function L Total , including: Evaluation item L of diagnostic accuracy of deep learning fault diagnosis model ce ; The source domain data similarity evaluation item L before and after the introduction of interference signals sim ; Before and after the introduction of the interference signal, the same source domain data is processed by the deep learning fault diagnosis model, and the similarity evaluation item L of the output feature dimension is dep ; The similarity evaluation item L of the bearing source domain data and the bearing target domain data after extracting features through the deep learning fault diagnosis model ang .

3. The method for diagnosing bearing faults of a lifting mechanism based on causal feature learning according to claim 2, characterized in that: Custom total loss function L ce , the specific formula is as follows: In the formula, α, β, γ are all hyperparameters, and N is the number of samples.

4. The method for diagnosing bearing faults of a lifting mechanism based on causal feature learning according to claim 2, characterized in that: Evaluation item L of diagnostic accuracy of deep learning fault diagnosis model ce , the Softmax classification layer adopts the cross entropy loss function, the specific formula is as follows: in, is the probability distribution calculated from the original output of the model by the Softmax function, y c is the true label, and R is the number of sample categories.

5. The method for diagnosing bearing faults of a hoisting mechanism based on causal feature learning according to claim 2, characterized in that: Source domain data similarity evaluation item L sim The specific acquisition method is as follows: Fourier transform of bearing source domain data; Performing perturbation operation on the bearing source domain data after Fourier transformation; Perform inverse Fourier transform on the bearing source domain data after the disturbance operation; The Pearson correlation coefficient is used to calculate the similarity between the original bearing source domain data and the bearing source domain data after Fourier inverse transformation, which is used as the source domain data similarity evaluation item L sim .

6. The method for diagnosing bearing faults of a hoisting mechanism based on causal feature learning according to claim 2, characterized in that: Output feature dimension similarity evaluation item L dep The specific acquisition method is as follows: Fourier transform of bearing source domain data; Performing perturbation operation on the bearing source domain data after Fourier transformation; Perform inverse Fourier transform on the bearing source domain data after the disturbance operation; The Ehrlich correlation coefficient is used to calculate the similarity of the output features of the fully connected layer of the deep learning fault diagnosis model for the same bearing source domain data before and after interference, which is used as the similarity evaluation item L of the output feature dimension. dep .

7. The method for diagnosing bearing faults of a hoisting mechanism based on causal feature learning according to claim 2, characterized in that: The similarity evaluation item L of the bearing source domain data and the bearing target domain data after extracting features through the deep learning fault diagnosis model ang The specific acquisition method is as follows: Extract feature vectors of bearing source domain data using deep learning fault diagnosis model; Extract feature vectors of bearing target domain data using deep learning fault diagnosis model; The angular distance metric is used to calculate the similarity of the two feature vectors, which is used as the similarity evaluation item L after the bearing source domain data and the bearing target domain data are extracted by the deep learning fault diagnosis model. ang .

8. The method for diagnosing bearing faults of a hoisting mechanism based on causal feature learning according to claim 1, characterized in that: The bearing source domain data is obtained through a bearing fault test and is labeled data; The bearing target domain data is acquired by detecting the actual working site of the lifting mechanism through sensors and is unlabeled data.

9. The method for diagnosing bearing faults of a hoisting mechanism based on causal feature learning according to claim 1, characterized in that: The bearing failure of the lifting mechanism includes: bearing inner ring failure and bearing outer ring failure.

Images