Multi-way valve element fault diagnosis method and device

Abstract

The invention provides a multi-way valve core fault diagnosis method and device, and the method is characterized in that the method comprises the following steps: S1, collecting a vibration signal, a flow signal and a pressure signal of a valve body; step S2-S3, generating a corresponding valve body vibration image, a flow image and a pressure image according to the valve body vibration signal, the flow signal and the pressure signal; s4, respectively inputting the valve body vibration image, the flow image and the pressure image into corresponding prediction models to respectively obtain corresponding prediction vectors; s5, performing calculation on each prediction model in combination with the prediction vector to obtain corresponding credibility and membership; and S6, performing fusion calculation according to all credibility and membership degrees to obtain fusion probabilities corresponding to all fault categories, and selecting the fault category corresponding to the maximum fusion probability as a multi-way valve core fault diagnosis result. In a word, the method can generate an accurate multi-way valve core fault diagnosis result.

Description

Technical Field

[0001] The invention belongs to the technical field of fault diagnosis, and in particular relates to a method and a device for diagnosing a multi-way valve core fault. Background Art

[0002] As one of the core components of the hydraulic system, the multi-way valve is a multifunctional integrated valve used to control the movement of multiple hydraulic actuators and to manipulate the movement of multiple actuators. The multi-way valve is suitable for various hydraulic systems due to its high performance, reliability and flexibility. It is mainly used in engineering machinery, lifting and transportation machinery and other mobile machinery that require the manipulation of multiple actuators. As a key component in the hydraulic system, the multi-way valve is particularly susceptible to the failure of components such as the valve core during operation. The performance of the multi-way valve determines the control accuracy and reliability of the hydraulic system. Failure to accurately diagnose and evaluate the type and severity of the fault will directly affect the operation of the equipment and even lead to major safety incidents and economic losses.

[0003] In order to solve this problem, many scholars have actively studied fault diagnosis methods in recent years. Relevant research has been conducted from the aspects of feature selection and deep learning. Deep learning is adaptive and has strong generalization. It can analyze the data collected from sensors and machines in real time and adaptively adjust the working process of equipment. Therefore, deep learning methods are becoming increasingly important in industrial applications.

[0004] However, existing deep learning-based fault diagnosis research shows some limitations. First, the performance of deep learning-based fault diagnosis models always relies on a large amount of high-quality labeled data. Since multi-way valves are often in a closed working environment in practice and faults are diverse, fault data is difficult to obtain, which limits the effectiveness and accuracy of model training. Secondly, the operating environment of multi-way valves will change over time, including equipment aging, changes in operating conditions, etc. Deep learning-based fault diagnosis models may need to be constantly retrained or adjusted when responding to dynamic environmental changes, increasing maintenance costs and complexity.

[0005] In summary, the existing deep learning-based multi-way valve fault diagnosis still has the problems of low accuracy and difficulty in sustainable use. Summary of the invention

[0006] The present invention is made to solve the above-mentioned problem, and aims to provide a method and device for diagnosing a multi-way valve core fault.

[0007] The present invention provides a multi-way valve core fault diagnosis method, which has the following characteristics and comprises the following steps: step S1, collecting the valve body vibration signal, flow signal and pressure signal when the multi-way valve is working; step S2, inputting the valve body vibration signal, flow signal and pressure signal together into an LSTM model to generate time-aligned valve body vibration time series data, flow time series data and pressure time series data; step S3, converting the valve body vibration time series data, flow time series data and pressure time series data into corresponding valve body vibration images, flow images and pressure images respectively through a Gram angle field; step S4, inputting the valve body vibration image, flow image and pressure image into corresponding prediction models respectively, and obtaining corresponding prediction vectors respectively; step S5, for each prediction model, respectively combining the corresponding prediction vector, respectively calculating the corresponding trust and membership of each prediction model; step S6, performing fusion calculation according to all trusts and memberships, obtaining the fusion probability corresponding to each fault category, and selecting the fault category corresponding to the maximum fusion probability as the multi-way valve core fault diagnosis result.

[0008] In the multi-way valve core fault diagnosis method provided by the present invention, it can also have the following characteristics: wherein, the prediction vector includes the prediction probability value of the corresponding prediction model for each of the fault categories, and the trust degree Bel (θ corresponding to the fault category θ of the i-th prediction model i ) is calculated as: P(D)=∫P(D|θ)P(θ)dθ, Where P(D|θ) is the ratio of the total number of accurate predictions of the i-th prediction model for fault category θ to the total number of predictions, and a and b are the lower and upper limits of the prediction probability value range of the i-th prediction model for fault category θ, respectively.

[0009] The multi-way valve core fault diagnosis method provided by the present invention may also have the following characteristics: wherein the membership degree C corresponding to the i-th prediction model is i The calculation expression is: F max =log(D), where α is the adjustment parameter, u i is the prediction vector output by the ith prediction model, c is the center point of the Gaussian function, σ is the standard deviation, and D is the total number of fault categories.

[0010] The multi-way valve core fault diagnosis method provided by the present invention may also have the following characteristics: wherein the fusion probability P corresponding to the fault category θ is com The calculation expression of (θ) is: m i (θ j )=Bel(θ j )×Ci , where n is the total number of prediction models, Bel(θ j ) is the trust degree of the i-th prediction model corresponding to the j-th fault category θ, C i is the membership degree corresponding to the i-th prediction model.

[0011] The multi-way valve core fault diagnosis method provided by the present invention may also have the following characteristics: wherein the prediction model is a neural network model, and the neural network model includes: a convolution layer, which is used to extract features of the input image to obtain a feature map; a pooling layer, which is used to select and filter features of the feature map to obtain a pooled output; and an output layer, which is used to obtain the predicted probability value corresponding to each fault category as a prediction vector through Softmax calculation based on the pooled output.

[0012] The multi-way valve core fault diagnosis method provided by the present invention may also have the following feature: wherein, the parameters of each convolution kernel in the convolution layer are shared and unchanged.

[0013] The multi-way valve core fault diagnosis method provided by the present invention may also have the following characteristics: wherein, in the pooling layer, a maximum pooling operation is performed on each feature map, and the calculation expression of the maximum pooling operation is: In the formula is the value of the (k, m) element of the i-th feature map of the L layer after the maximum pooling operation, and L and W are the length and width of the pooling window respectively.

[0014] The multi-way valve core fault diagnosis method provided by the present invention may also have the following characteristics: wherein, in the output layer, the calculation expression of the predicted probability value is: Where p(y j ) is the predicted probability value corresponding to the jth fault category, y j is the output of the jth neuron in the output layer, and q is the total number of fault categories.

[0015] The multi-way valve core fault diagnosis method provided by the present invention may also have the following features: wherein, in step S1, a pressure signal is collected by a pressure sensor, a flow signal is collected by a flow sensor, and a valve body vibration signal is collected by an acceleration sensor. The pressure sensor, the flow sensor and the acceleration sensor are all attached to the outer surface of the valve body of the multi-way valve and are close to the oil outlet and the oil inlet of the multi-way valve.

[0016] The present invention also provides a multi-way valve core fault diagnosis device, which has the following characteristics, including: a signal acquisition module, which is used to collect the valve body vibration signal, flow signal and pressure signal when the multi-way valve is working; a timing generation module, which is used to input the valve body vibration signal, flow signal and pressure signal together into the LSTM model to generate time-aligned valve body vibration time series data, flow time series data and pressure time series data; an image conversion module, which is used to convert the valve body vibration time series data, flow time series data and pressure time series data into corresponding valve body vibration images, flow images and pressure images respectively through the Gram angle field; a prediction module, which includes multiple prediction models, which are used to input the valve body vibration image, flow image and pressure image into the corresponding prediction models respectively, and obtain corresponding prediction vectors respectively; a calculation module, which is used to calculate the corresponding trust and membership of each prediction model respectively in combination with the corresponding prediction vector; a fusion module, which is used to perform fusion calculation based on all trust and membership to obtain the fusion probability corresponding to each fault category, and select the fault category corresponding to the maximum fusion probability as the multi-way valve core fault diagnosis result.

[0017] Functions and Effects of the Invention

[0018] According to the multi-way valve core fault diagnosis method and device involved in the present invention, because, first, the Gram angle field is used to convert one-dimensional data into a two-dimensional image, and then a neural network model with fewer parameters and calculations, that is, a prediction model, is used for processing, which can effectively improve the accuracy of fault diagnosis and is more efficient than complex models; second, the prediction results of multiple different prediction models are fused by combining the prior information of Bayesian reasoning, fuzzy logic theory and DS theory to generate a fusion probability; third, the impact of conflicts is effectively controlled through the prior information of Bayesian decision-making, especially when the conflict between multiple evidence sources is more serious, the Bayesian method can alleviate the conflict by weighting the evidence and correcting the prior distribution. Therefore, the multi-way valve core fault diagnosis method and device of the present invention can generate accurate multi-way valve core fault diagnosis results. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a block diagram of a multi-way valve core fault diagnosis device in an embodiment of the present invention;

[0020] Figure 2 is a block diagram of a neural network model in an embodiment of the present invention;

[0021] Figure 3 It is a flow chart of a method for diagnosing a multi-way valve core fault in an embodiment of the present invention. DETAILED DESCRIPTION

[0022] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the following embodiments and the accompanying drawings specifically illustrate the multi-way valve core fault diagnosis method and device of the present invention.

[0023] Figure 1 It is a block diagram of a multi-way valve core fault diagnosis device in an embodiment of the present invention.

[0024] like Figure 1 As shown, the multi-way valve core fault diagnosis device 100 includes a signal acquisition module 11, a timing generation module 12, an image conversion module 13, a prediction module 14, a calculation module 15, a fusion module 16 and a master control module 17 for controlling the operation of the above modules.

[0025] The signal acquisition module 11 is used to acquire the valve body vibration signal, flow signal and pressure signal when the multi-way valve is working.

[0026] Among them, the signal acquisition module 11 includes a pressure sensor, a flow sensor and an acceleration sensor. The signal acquisition module 11 collects pressure signals through the pressure sensor, collects flow signals through the flow sensor, and collects valve body vibration signals through the acceleration sensor. The pressure sensor, flow sensor and acceleration sensor are all attached to the outer surface of the valve body of the multi-way valve, and are close to the oil outlet and oil inlet of the multi-way valve. In this embodiment, the pressure sensor, flow sensor and acceleration sensor are all attached to the multi-way valve by magnetic attraction.

[0027] The time series generation module 12 is used to input the valve body vibration signal, flow signal and pressure signal into the LSTM model to generate time-aligned valve body vibration time series data, flow time series data and pressure time series data.

[0028] The image conversion module 13 is used to convert the valve body vibration time series data, flow rate time series data and pressure time series data into corresponding valve body vibration images, flow rate images and pressure images respectively through the Gram angle field.

[0029] In this embodiment, the Gram angular field generates an image by converting the values ​​of one-dimensional time series data into angles in polar coordinates and calculating the cosine values ​​between these angles. Compared with traditional time-frequency analysis methods such as short-time Fourier transform and wavelet transform, the Gram angular field can better ensure data integrity and express time dependence. In addition, compared with other image coding methods, the Gram angular field provides an intuitive and effective way to visualize and analyze time series data through the angle method, effectively capturing the periodicity and symmetry characteristics of the data.

[0030] The prediction module 14 includes multiple prediction models, which are used to input the valve body vibration image, flow image and pressure image into corresponding prediction models respectively to obtain corresponding prediction vectors. The prediction vector includes the prediction probability value of each fault category of the corresponding prediction model. The prediction model is a neural network model.

[0031] Figure 2 It is a block diagram of a neural network model in an embodiment of the present invention.

[0032] like Figure 2 As shown, the neural network model 200 includes a convolution layer 21, a pooling layer 22 and an output layer 23.

[0033] The convolution layer 21 is used to extract features from the input image to obtain a feature map.

[0034] Among them, the parameters of each convolution kernel in the convolution layer 21 are shared and unchanged, which can not only reduce the training parameters of the model, but also fully learn the local features contained in the data.

[0035] In this embodiment, by increasing the number of convolution kernels, each convolution kernel performs a convolution operation on the input image to generate a feature map, which represents the features extracted by the convolution kernel from the input image. Different feature maps are thus formed, and these different feature maps are combined as the output of the convolution layer, so that the input data has a certain degree of displacement invariance and scaling invariance.

[0036] The pooling layer 22 is used to perform feature selection and filtering on the feature map to obtain a pooled output. In this embodiment, the pooling layer 22 reduces the resolution of the extracted feature map and continuously reduces the spatial size of the data, so the number of parameters and the amount of calculation will also decrease, which can control overfitting to a certain extent. The pooling layer 22 can perform average pooling or maximum pooling operations on the feature map as a pooling operation.

[0037] In the pooling layer 22 of this embodiment, a maximum pooling operation is performed on each feature map. The calculation expression of the maximum pooling operation is:

[0038]

[0039] In the formula is the value of the (k, m) element of the i-th feature map of the L layer after the maximum pooling operation, and L and W are the length and width of the pooling window respectively.

[0040] The output layer 23 is used to obtain the predicted probability value corresponding to each fault category as a prediction vector through Softmax calculation based on the pooled output.

[0041] Among them, in the output layer 23, the calculation expression of the predicted probability value is:

[0042]

[0043] Where p(y j ) is the predicted probability value corresponding to the jth fault category, y j is the output of the jth neuron in the output layer, and q is the total number of fault categories.

[0044] Each prediction model of this embodiment is obtained by training with existing training data. The existing training data includes the flow signal and pressure signal obtained by collecting the pressure and flow signals of each working link of the multi-way valve under different working conditions using a pressure sensor and a flow sensor, and the valve body vibration signal obtained by collecting the valve body vibration signal during the working process of the multi-way valve using an acceleration sensor. In this embodiment, different working conditions include different loads and valve core displacement frequencies, stable operation, and valve core stuck states.

[0045] The calculation module 15 is used to calculate the trust and membership of each prediction model in combination with the corresponding prediction vector.

[0046] Among them, the trust degree Bel(θ corresponding to the fault category θ of the i-th prediction model is i ) is calculated as:

[0047]

[0048] P(D)=∫P(D|θ)P(θ)dθ,

[0049]

[0050] Where P(D|θ) is the ratio of the total number of accurate predictions of the i-th prediction model for fault category θ to the total number of predictions, and a and b are the lower and upper limits of the prediction probability value range of the i-th prediction model for fault category θ, respectively.

[0051] When the prediction model of this embodiment processes real-time data, the new observation data D t The probability distribution of each node in the Bayesian network will be continuously updated. i When the state of changes, that is, the new data D t When , the Bayesian network recalculates its posterior probability based on the new evidence. There are multiple observation data D at time t t , update the node θ by the following recursive formula i Status:

[0052]

[0053] The above process continues each time new data is received until the probability distribution of all nodes in the network is updated.

[0054] In this embodiment, the prediction results and actual results of the prediction model are used as historical data to update the prior information P(D |θ), that is, the P(D |θ) of each prediction model is continuously and dynamically updated. Therefore, the prediction model does not need to be retrained or adjusted in the dynamic environment changes, but the trust of each prediction model is fine-tuned, that is, the prediction model is dynamically weighted.

[0055] The membership degree C corresponding to the i-th prediction model i The calculation expression is:

[0056]

[0057]

[0058]

[0059] F max =log(D),

[0060] Where α is the adjustment parameter used to control the slope of the mapping, u i is the prediction vector output by the ith prediction model, c is the center point of the Gaussian function, σ is the standard deviation U→[0, 1], and D is the total number of fault categories. In this embodiment, 11 fault categories are set when training the prediction model, so D=11 in this embodiment.

[0061] In this embodiment, the membership function μ F (u) is the Gaussian membership function, which is smooth and continuous in the entire definition domain. This feature enables the fuzzy system based on the Gaussian membership function to handle small changes between similar results more smoothly, avoiding changes caused by the discontinuity of the membership function, thereby improving the stability and reliability of fault diagnosis.

[0062] The fusion module 16 is used to perform fusion calculation according to all trust degrees and membership degrees to obtain the fusion probability corresponding to each fault category, and select the fault category corresponding to the maximum fusion probability as the multi-way valve core fault diagnosis result.

[0063] Among them, the fusion probability P corresponding to the fault category θ com The calculation expression of (θ) is:

[0064]

[0065]

[0066] m i (θ j)=Bel(θ j )×C i ,

[0067] Where n is the total number of prediction models, Bel(θ j ) is the trust degree of the i-th prediction model corresponding to the j-th fault category θ, C i is the membership degree corresponding to the i-th prediction model.

[0068] The master control module 17 stores a control program for controlling the operation of each module.

[0069] The following describes the process of a method for diagnosing a multi-way valve core fault using the multi-way valve core fault diagnosis device 100 in conjunction with the accompanying drawings.

[0070] Figure 3 It is a flow chart of a method for diagnosing a multi-way valve core fault in an embodiment of the present invention.

[0071] like Figure 3 As shown, the multi-way valve core fault diagnosis method includes the following steps:

[0072] Step S1, using the signal acquisition module 11 to acquire the valve body vibration signal, flow signal and pressure signal when the multi-way valve is working.

[0073] Step S2, using the timing generation module 12 to input the valve body vibration signal, flow signal and pressure signal into the LSTM model to generate time-aligned valve body vibration time series data, flow time series data and pressure time series data.

[0074] Step S3, using the image conversion module 13 to convert the valve body vibration time series data, flow rate time series data and pressure time series data into corresponding valve body vibration images, flow rate images and pressure images respectively through the Gram angle field.

[0075] Step S4: using the prediction module 14 to input the valve body vibration image, the flow image and the pressure image into corresponding prediction models respectively, and obtain corresponding prediction vectors respectively.

[0076] Step S5: The calculation module 15 is used to calculate the trust and membership of each prediction model in combination with the corresponding prediction vector.

[0077] Step S6, using the fusion module 16 to perform fusion calculation according to all the trusts and memberships to obtain the fusion probability corresponding to each fault category, and selecting the fault category corresponding to the maximum fusion probability as the multi-way valve core fault diagnosis result.

[0078] Functions and Effects of the Embodiments

[0079] According to the multi-way valve core fault diagnosis method and device involved in this embodiment, first, the Gram angle field is used to convert one-dimensional data into a two-dimensional image, and then a neural network model with fewer parameters and calculations, namely a prediction model, is used for processing, which can effectively improve the accuracy of fault diagnosis and is more efficient than complex models; second, the prediction results of multiple different prediction models are fused by combining the prior information of Bayesian reasoning, fuzzy logic theory and DS theory to generate a fusion probability; third, the impact of conflicts is effectively controlled through the prior information of Bayesian decision-making, especially when the conflict between multiple evidence sources is more serious, the Bayesian method can alleviate the conflict by weighting the evidence and correcting the prior distribution. In short, this method can generate accurate multi-way valve core fault diagnosis results.

[0080] Furthermore, the prediction results of the latest prediction model and the actual fault update prior information can be used to perform real-time dynamic weighting to improve the accuracy of multi-way valve core fault diagnosis in a dynamic environment.

[0081] Those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for diagnosing a multi-way valve core fault, characterized in that: The following steps are involved: Step S1, collecting a valve body vibration signal, a flow signal and a pressure signal when the multi-way valve is working; Step S2, inputting the valve body vibration signal, the flow signal and the pressure signal together into an LSTM model to generate time-aligned valve body vibration time series data, flow time series data and pressure time series data; Step S3, converting the valve body vibration time series data, the flow rate time series data and the pressure time series data into corresponding valve body vibration images, flow rate images and pressure images respectively through the Gram angle field; Step S4, inputting the valve body vibration image, the flow image and the pressure image into corresponding prediction models respectively, and obtaining corresponding prediction vectors respectively; Step S5, for each prediction model, respectively, in combination with the corresponding prediction vector, respectively, calculate the trust and membership corresponding to each prediction model; Step S6, performing fusion calculation based on all the trust degrees and the membership degrees to obtain the fusion probability corresponding to each fault category, and selecting the fault category corresponding to the largest fusion probability as the multi-way valve core fault diagnosis result.

2. The multi-way valve core fault diagnosis method according to claim 1 is characterized in that: in, The prediction vector includes the prediction probability value of the corresponding prediction model for each of the fault categories. The confidence level Bel(θ corresponding to the fault category θ of the i-th prediction model i ) is calculated as: P(D)=∫P(D|θ)P(θ)dθ, Where P(D|θ) is the ratio of the total number of accurate predictions of the i-th prediction model for fault category θ to the total number of predictions, and a and b are the lower and upper limits of the prediction probability value range of the i-th prediction model for fault category θ, respectively.

3. The multi-way valve core fault diagnosis method according to claim 1 is characterized in that: in, The membership degree C corresponding to the i-th prediction model i The calculation expression is: F max =log(D), Where α is the adjustment parameter, u i is the prediction vector output by the ith prediction model, c is the center point of the Gaussian function, σ is the standard deviation, and D is the total number of fault categories.

4. The method for diagnosing a multi-way valve core fault according to claim 1, characterized in that: in, The fusion probability P corresponding to the fault category θ con The calculation expression of (θ) is: m i (i j )=Bel(θ j )×C i , Where n is the total number of prediction models, Bel(θ j ) is the trust degree of the i-th prediction model corresponding to the j-th fault category θ, C i is the membership degree corresponding to the i-th prediction model.

5. The multi-way valve core fault diagnosis method according to claim 1, Features: Wherein, the prediction model is a neural network model, The neural network model includes: The convolution layer is used to extract features from the input image and obtain a feature map; A pooling layer, used to perform feature selection and filtering on the feature map to obtain a pooled output; The output layer is used to obtain the predicted probability value corresponding to each of the fault categories as a prediction vector through Softmax calculation according to the pooled output.

6. The method for diagnosing a multi-way valve core fault according to claim 5, characterized in that: in, In the convolution layer, the parameters of the convolution kernels in each layer are shared and unchanged.

7. The method for diagnosing a multi-way valve core fault according to claim 5, characterized in that: in, In the pooling layer, a maximum pooling operation is performed on each of the feature maps. The calculation expression of the maximum pooling operation is: In the formula is the value of the (k,m) element of the i-th feature map of the L layer after the maximum pooling operation, and L and W are the length and width of the pooling window respectively.

8. The method for diagnosing a multi-way valve core fault according to claim 5, characterized in that: in, In the output layer, the calculation expression of the predicted probability value is: Where p(y j ) is the predicted probability value corresponding to the jth fault category, y j is the output of the jth neuron in the output layer, and q is the total number of fault categories.

9. The multi-way valve core fault diagnosis method according to claim 1, characterized in that: in, In step S1, the pressure signal is collected by a pressure sensor. The flow signal is collected by a flow sensor, The valve body vibration signal is collected by an acceleration sensor. The pressure sensor, the flow sensor and the acceleration sensor are all attached to the outer surface of the valve body of the multi-way valve and are close to the oil outlet and the oil inlet of the multi-way valve.

10. A multi-way valve core fault diagnosis device, characterized in that: include: A signal acquisition module is used to collect valve body vibration signals, flow signals and pressure signals when the multi-way valve is working; A time series generation module, used for inputting the valve body vibration signal, the flow signal and the pressure signal together into an LSTM model to generate time-aligned valve body vibration time series data, flow time series data and pressure time series data; An image conversion module, used to convert the valve body vibration time series data, the flow rate time series data and the pressure time series data into corresponding valve body vibration images, flow rate images and pressure images respectively through a Gram angle field; A prediction module, comprising a plurality of prediction models, for inputting the valve body vibration image, the flow image and the pressure image into corresponding prediction models respectively to obtain corresponding prediction vectors respectively; A calculation module, used for calculating the trust and membership of each prediction model in combination with the corresponding prediction vector; The fusion module is used to perform fusion calculation according to all the trust degrees and the membership degrees to obtain the fusion probability corresponding to each fault category, and select the fault category corresponding to the largest fusion probability as the multi-way valve core fault diagnosis result.

Images