Hydraulic cylinder fault diagnosis model training method and hydraulic cylinder fault diagnosis method

Abstract

The embodiment of the application provides a hydraulic cylinder fault diagnosis model training method and a hydraulic cylinder fault diagnosis method, and belongs to the field of hydraulic fault diagnosis. The method comprises the following steps: acquiring data signals of a plurality of sensors, wherein the plurality of sensors are used to collect a plurality of types of state data of the hydraulic cylinder; performing feature extraction and fault type marking on the data signals to obtain a feature vector set and a fault type label, wherein the fault type label comprises a corresponding relationship between a single fault and a related feature vector, and a corresponding relationship between a coupling fault and a related feature vector; inputting the feature vector set and the fault type label into a to-be-trained model to perform model training, so as to obtain a fault diagnosis model; and using the fault diagnosis model to perform fault classification of the hydraulic cylinder according to acquired fault data of the plurality of sensors. Through the method, the problem that existing hydraulic cylinder fault diagnosis technology cannot diagnose the coupling fault of the hydraulic cylinder can be overcome.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of hydraulic fault diagnosis, in particular to a hydraulic cylinder fault diagnosis model training method and a hydraulic cylinder fault diagnosis method. BACKGROUND

[0002] As a key execution component of engineering machinery equipment, the hydraulic cylinder has a complex working environment, working condition and application scenario, so it often has not only a single fault, but also a variety of concurrent fault coupling phenomena. At present, the data-driven hydraulic cylinder fault diagnosis technology only studies the single specific fault type, and the field of self-diagnosis of the diversity of the coupling fault of the hydraulic cylinder is still a blank. SUMMARY

[0003] The purpose of the present application is to overcome the problem that the existing hydraulic cylinder fault diagnosis technology cannot diagnose the coupling fault of the hydraulic cylinder, and to provide a hydraulic cylinder fault diagnosis model training method, characterized by comprising:

[0004] Obtaining data signals of a plurality of sensors, wherein the plurality of sensors are used to collect a plurality of state data of the hydraulic cylinder;

[0005] Performing feature extraction and fault type labeling on the data signals to obtain a feature vector set and a fault type label, wherein the fault type label includes a corresponding relationship between a single fault and a related feature vector, and a corresponding relationship between a coupling fault and a related feature vector;

[0006] Inputting the feature vector set and the fault type label into a to-be-trained model for model training to obtain a fault diagnosis model.

[0007] In an embodiment of the present application, the to-be-trained model includes a pre-trained model and a top-level classifier, and the feature vector set and the fault type label are input into the to-be-trained model for model training to obtain the fault diagnosis model, comprising:

[0008] Step S310: inputting the feature vector set into a pre-trained model composed of a plurality of layers of restricted Boltzmann machines for pre-training to obtain a second feature vector set;

[0009] Step S320: inputting the second feature vector set and the fault type label into the top-level classifier for iterative training until a preset iterative training completion condition is met;

[0010] Step S330: determining whether the classification accuracy of the to-be-trained model meets a preset first accuracy requirement and whether the error curve of the to-be-trained model converges according to the classification result output by the top-level classifier after completing the iterative training;

[0011] Step S340: in the case that the classification accuracy meets the first accuracy requirement and the error curve converges, the training of the to-be-trained model is completed, and a fault diagnosis model is obtained;

[0012] In the case that the classification accuracy does not meet the first accuracy requirement or the error curve does not converge, the type of the top-level classifier is replaced, and steps S320 and S330 are repeated.

[0013] In an embodiment of the present application, step 310 comprises:

[0014] The feature vector set is input into the pre-trained model;

[0015] It is judged whether the error between the reconstructed vector set output by the pre-trained model and the feature vector set meets a preset second accuracy requirement;

[0016] In the case that the error between the reconstructed vector set and the feature vector set does not meet the preset second accuracy requirement, the preset learning parameter of the pre-trained model is adjusted, and the pre-training is performed again;

[0017] In the case that the error between the reconstructed vector set and the feature vector set meets the preset second accuracy requirement, the reconstructed vector set is output as a second feature vector set.

[0018] In an embodiment of the present application, the top-level classifier further comprises an endocrine gland for secreting a hormone, and the hormone is used for adjusting the input of neurons in the to-be-trained model, and step S320 comprises:

[0019] The second feature vector set and the fault type label are input into the top-level classifier;

[0020] An error value is determined according to a preset error function, a classification result output by the top-level classifier and the fault type label;

[0021] The concentration of the hormone is adjusted according to a preset concentration decay function and the error value;

[0022] The weight and skewness value of the input of each neuron in the top-level classifier and the weight and skewness value of the input of each neuron in the pre-trained model are adjusted according to the error value, until an iterative training completion condition is met.

[0023] In an embodiment of the present application, the data signal comprises a pressure signal, and the feature extraction and fault type labeling of the data signal are performed to obtain the feature vector set and the fault type label, comprising:

[0024] The pressure signal is denoised and empirically mode decomposed to obtain a plurality of intrinsic mode functions and a residual error;

[0025] determining an inherent modal energy value, an inherent modal energy entropy, a dimensional parameter and a non-dimensional parameter of each inherent modal function to obtain time domain features of the pressure signal;

[0026] performing a Hilbert transform on each inherent modal function, and determining a mean value, a root mean square and a root mean square of a transient frequency of a transient amplitude of each transformed inherent modal function to obtain time-frequency domain features of the pressure signal;

[0027] obtaining a feature vector set of the pressure signal by using the time domain features and the time-frequency domain features of the pressure signal;

[0028] performing corresponding fault type marking on the obtained time domain features and time-frequency domain features to obtain a fault type label.

[0029] In an embodiment of the present application, the data signals include a piston rod displacement signal, a vibration signal and an acoustic emission signal, and feature extraction and fault type marking are performed on the data signals to obtain a feature vector set and a fault type label, including:

[0030] converting the piston rod displacement signal into a piston rod acceleration signal;

[0031] performing frame division on the piston rod acceleration signal, the vibration signal and the acoustic emission signal, and calculating a mean value, a root mean square, a skewness value, a peak factor, a pulse factor, a waveform factor, a margin factor and a kurtosis factor of each frame signal to obtain time-frequency domain features of the piston rod acceleration signal, the vibration signal and the acoustic emission signal respectively;

[0032] performing CQCC feature extraction on each frame signal;

[0033] obtaining a feature vector set of the piston rod displacement signal, the vibration signal and the acoustic emission signal respectively by using the time-frequency domain features and the CQCC features of the piston rod displacement signal, the vibration signal and the acoustic emission signal respectively;

[0034] performing corresponding fault type marking on the obtained time-frequency domain features and CQCC features to obtain a fault type label.

[0035] In an embodiment of the present application, the data signals include a flow signal and a temperature signal, and feature extraction and fault type marking are performed on the data signals to obtain a feature vector set and a fault type label, including:

[0036] performing corresponding fault type marking on the flow signal and the temperature signal to obtain a fault type label.

[0037] The second aspect of the present application provides a hydraulic cylinder fault diagnosis method, including:

[0038] obtaining fault data, wherein the fault data includes data signals of a plurality of sensors;

[0039] extracting features from the fault data to obtain a fault feature vector;

[0040] inputting the fault feature vector into a fault diagnosis model to obtain a fault category corresponding to the fault data, wherein the fault diagnosis model is obtained by the training method of the hydraulic cylinder fault diagnosis model as described above.

[0041] The third aspect of the present application provides a training device of a hydraulic cylinder fault diagnosis model, comprising:

[0042] a data acquisition unit configured to acquire data signals of a plurality of sensors, wherein the plurality of sensors are configured to collect a plurality of state data of the hydraulic cylinder;

[0043] a feature extraction and labeling unit configured to extract features from the data signals and label fault types to obtain a feature vector set and a fault type label, wherein the fault type label comprises a corresponding relationship between a single fault and a related feature vector, and a corresponding relationship between a coupling fault and a related feature vector;

[0044] a model training unit configured to input the feature vector set and the fault type label into a to-be-trained model to perform model training to obtain a fault diagnosis model.

[0045] The fourth aspect of the present application provides a hydraulic cylinder fault diagnosis device, comprising:

[0046] a fault detection unit configured to acquire fault data, wherein the fault data comprises data signals of a plurality of sensors;

[0047] a fault feature extraction unit configured to extract features from the fault data to obtain a fault feature vector;

[0048] a fault classification unit configured to input the fault feature vector into a fault diagnosis model to obtain a specific fault category, wherein the fault diagnosis model is obtained by the training method of the hydraulic cylinder fault diagnosis model as described above.

[0049] The fifth aspect of the present application provides an electronic device comprising a processor and a memory, wherein the memory stores machine executable instructions capable of being executed by the processor, and the processor is capable of executing the machine executable instructions to implement the training method of the hydraulic cylinder fault diagnosis model or the hydraulic cylinder fault diagnosis method as described above.

[0050] The sixth aspect of the present application provides a machine readable storage medium, wherein the machine readable storage medium stores instructions, and the instructions are executed by a processor to cause the processor to implement the training method of the hydraulic cylinder fault diagnosis model or the hydraulic cylinder fault diagnosis method as described above.

[0051] By the technical solution, multiple sensors are adopted to collect multiple physical source data signals representing the state of the hydraulic cylinder, feature extraction and fault type marking are performed on the data signals, the extracted feature vectors are used to train a fault diagnosis model, and in the subsequent fault diagnosis process, only real-time fault data collected by the sensors need to be input into the trained fault diagnosis model, so that the coupling fault and single fault that may occur simultaneously in the hydraulic cylinder are classified.

[0052] Other features and advantages of the embodiments of the present application will be described in detail in the following detailed description. BRIEF DESCRIPTION OF DRAWINGS

[0053] The accompanying drawings are included to provide a further understanding of the embodiments of the present application, and constitute a part of the specification, and are used to explain the embodiments of the present application together with the following detailed description, but do not constitute a limitation of the embodiments of the present application. In the drawings:

[0054] Figure 1 A flowchart schematically showing a training method of a hydraulic cylinder fault diagnosis model according to an embodiment of the present application is shown;

[0055] Figure 2 A flowchart schematically showing an iterative training of a training method of a hydraulic cylinder fault diagnosis model according to an embodiment of the present application is shown;

[0056] Figure 3 A flowchart schematically showing a hydraulic cylinder fault diagnosis method according to an embodiment of the present application is shown;

[0057] Figure 4 A structural diagram schematically showing a deep belief neural network according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0058] The specific embodiments of the present application are described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only used to illustrate and explain the present application, and are not used to limit the present application.

[0059] It should be noted that if the present application involves directional indications (such as up, down, left, right, front, back, etc.), the directional indications are only used to explain the relative positional relationship, reversing condition, etc. between the components in a certain specific posture (as shown in the drawings), if the specific posture changes, the directional indications also change accordingly.

[0060] In addition, if the description of "first", "second" and the like is involved in the embodiments of the present application, the description of "first", "second" and the like is only for the purpose of description, and cannot be understood as indicating or implying the relative importance of the indicated technical features or implicitly indicating the number of the indicated technical features. Therefore, the features limited by "first", "second" can be explicitly or implicitly included at least one of the features. In addition, the technical solutions of various embodiments can be combined with each other, but it must be based on the realization of ordinary skilled in the art, when the combination of technical solutions appears contradictory or unachievable, it should be considered that the combination of technical solutions does not exist, nor in the protection scope claimed by the present application.

[0061] Figure 1 The flowchart of the training method of the hydraulic cylinder fault diagnosis model according to the embodiments of the present application is schematically shown, as shown in Figure 1 In one embodiment of the present application, a training method of a hydraulic cylinder fault diagnosis model is provided, which can include steps S100, S200 and S300.

[0062] Step S100: acquiring data signals of multiple sensors, wherein the multiple sensors are used to collect multiple types of state data of the hydraulic cylinder.

[0063] The data of the training set of the hydraulic cylinder fault diagnosis model is derived from multiple different types of sensors, the types of sensors can include pressure sensors, vibration sensors, temperature sensors, flow sensors, acoustic emission sensors and displacement sensors, which are used to collect multiple types of state data of the hydraulic cylinder, and also used to monitor the health status of the hydraulic cylinder in real time, and the processor uses the data signals of the sensors containing multiple types of state data for model training. Specifically, the pressure sensor is installed at the inlet and outlet of the hydraulic cylinder to monitor the pressure of the rod cavity and the pressure of the rodless cavity; the vibration sensor is installed on the cylinder, the end cover and the piston rod respectively to monitor the vibration of each part; the flow sensor is connected to the outlet pipe of the hydraulic cylinder; the displacement sensor is vertically installed on the top of the piston rod; the temperature sensor is connected to the inlet and outlet oil pipe of the hydraulic cylinder; the acoustic emission sensor is installed on the outer cylinder wall of the guide sleeve part to monitor the abnormal condition of the piston rod and the guide sleeve. In order to provide data support for the subsequent training model, for the common faults of the hydraulic cylinder, the data signals include the data signals corresponding to the single fault of the hydraulic cylinder and the signals corresponding to the coupled fault.

[0064] Exemplarily, the single failure of the hydraulic cylinder can include: hydraulic cylinder stretch (minor, serious), hydraulic cylinder seal ring damage (minor, serious), and piston rod surface being eroded (minor, serious). The above three types of single failures cause internal leakage of the hydraulic cylinder, and the corresponding data signals are: pressure signals and flow signals of two cavities of the hydraulic cylinder; the single failure of the hydraulic cylinder can also include: cavitation phenomenon of hydraulic oil, causing the hydraulic cylinder to creep, temperature to be abnormal, and the corresponding data signals are: displacement signals, vibration signals, acoustic emission signals and temperature signals; the single failure of the hydraulic cylinder can also include: axial offset of the hydraulic cylinder guide sleeve, hydraulic cylinder piston loosening, slight bending of the cylinder barrel, and loosening of the end cover nut or screw. The above three types of single failures cause high motion noise of the hydraulic cylinder, abnormal vibration frequency, cylinder body shaking, and creeping, and the corresponding data signals are: displacement signals, vibration signals and acoustic emission signals. The coupled failure of the hydraulic cylinder can be a random combination of the above single failures, such as hydraulic cylinder stretch, seal ring damage, and piston being eroded, or hydraulic cylinder stretch and cavitation phenomenon. The phenomenon caused by the coupled failure is a combination of the phenomena caused by the single failures that constitute the coupled failure, and the corresponding data signals are also a combination of the data signals corresponding to the single failures that constitute the coupled failure.

[0065] Step S200: feature extraction and failure type labeling are performed on the data signals to obtain a feature vector set and a failure type label, wherein the failure type label includes a corresponding relationship between a single failure and a related feature vector, and a corresponding relationship between a coupled failure and a related feature vector.

[0066] The data signals of multiple physical sources need to be analyzed in a data-driven manner to establish multiple-source collaborative failure features, and the feature vectors are labeled with failure types to obtain failure labels for supervised model training.

[0067] It can be understood that different data signals use different feature extraction and failure type labeling methods. The following will introduce the corresponding feature extraction and failure type labeling methods for different data signals.

[0068] In an embodiment of the present application, the data signals include pressure signals, and the feature extraction and failure type labeling are performed on the data signals to obtain a feature vector set and a failure type label, including:

[0069] The pressure signals are denoised and empirically mode decomposed to obtain a plurality of intrinsic mode functions and a residual;

[0070] The intrinsic mode energy value, the intrinsic mode energy entropy, the dimensional parameter and the dimensionless parameter of each intrinsic mode function are determined to obtain the time domain features of the pressure signals;

[0071] performing Hilbert transform on each intrinsic mode function, and determining the mean value, root mean square and root mean square of transient frequency of the transient amplitude of each intrinsic mode function after the transform to obtain time-frequency domain features of the pressure signal;

[0072] Using the time domain features and the time-frequency domain features of the pressure signal, a feature vector set of the pressure signal is obtained.

[0073] The obtained time domain features and time-frequency domain features are marked with corresponding fault types to obtain fault type labels.

[0074] Because there are multiple interference sources in the hydraulic system, the pressure signal in the data signal of the multiple physical sources needs to be denoised by wavelet to exclude interference. The pressure signal after denoising is decomposed by empirical mode decomposition (EMD, Empirical Mode Decomposition). As a time-frequency domain processing method, compared with wavelet analysis which is often applied in time-frequency domain processing scenarios, the most significant feature of EMD is that it overcomes the problem of non-adaptive base function. Wavelet analysis needs to select a certain wavelet base, and the selection of the wavelet base greatly affects the result of the entire wavelet analysis. Once the wavelet base is determined, it cannot be changed in the entire analysis process. Even if the wavelet base is the best in the global, it may not be the best in some local. Therefore, the base function of wavelet analysis lacks adaptability. EMD does not need to do pre-analysis and research on an unknown signal, and can directly start decomposition. This method will automatically divide according to some fixed mode, without human intervention.

[0075] Exemplarily, after EMD, 8 intrinsic mode functions (IMF) and a residual are obtained. The intrinsic mode energy value, intrinsic mode energy entropy, dimensional parameters (mean value, peak value, standard deviation, rectified mean value, variance, root mean square) and non-dimensional parameters (skewness value, peak factor, pulse factor, waveform factor, margin factor and kurtosis factor) of the 8 intrinsic mode functions are calculated, and 12 time domain indexes (two groups of pressure signals of the two cavities of the hydraulic cylinder form 192-dimensional time domain features) are obtained, that is, the time domain features of the pressure signal are obtained. Then, Hilbert transform is performed on the 8 intrinsic mode functions, and the mean value, root mean square and root mean square of transient frequency of the transient amplitude of each intrinsic mode function are calculated (two groups of pressure signals of the two cavities of the hydraulic cylinder form 48-dimensional time-frequency domain features), that is, the time-frequency domain features of the pressure signal are obtained. The time domain features and the time-frequency domain features of the pressure signal constitute a feature vector set of the pressure signal, and the time domain features and the time-frequency domain features in the feature vector set are marked with corresponding fault types to obtain fault type labels.

[0076] In an embodiment of the present application, the data signals include a piston rod displacement signal, a vibration signal and an acoustic emission signal, feature extraction and fault type labeling are performed on the data signals to obtain a feature vector set and a fault type label, including:

[0077] The piston rod displacement signal is converted into a piston rod acceleration signal;

[0078] The piston rod acceleration signal, the vibration signal and the acoustic emission signal are framed, and the mean value, the root mean square, the skewness value, the peak factor, the pulse factor, the waveform factor, the margin factor and the kurtosis factor of each frame signal are calculated to obtain the time-frequency domain features of the piston rod acceleration signal, the vibration signal and the acoustic emission signal respectively;

[0079] CQCC feature extraction is performed on each frame signal;

[0080] The time-frequency domain features and the CQCC features of the piston rod displacement signal, the vibration signal and the acoustic emission signal are used to obtain the feature vector set of the piston rod displacement signal, the vibration signal and the acoustic emission signal respectively;

[0081] The obtained time-frequency domain features and the CQCC features are labeled with corresponding fault types to obtain the fault type label.

[0082] The piston rod displacement signal, the vibration signal and the acoustic emission signal are used as the same group of data features for feature extraction. First, the piston rod displacement signal is converted into a piston rod acceleration signal, the piston rod acceleration signal, the vibration signal and the acoustic emission signal are framed, and the mean value, the root mean square, the skewness value, the peak factor, the pulse factor, the waveform factor, the margin factor and the kurtosis factor of each frame are calculated (5 sensor data form 40-dimensional time-frequency domain features), that is, the time-frequency domain features of the piston rod acceleration signal, the vibration signal and the acoustic emission signal are obtained; then each frame signal x(n) is windowed, and then constant Q transform (CQT, Constant Q transform) is performed, the time domain signal x(n) is converted into a frequency domain signal X CQ (k); the frequency signal is represented in complex form, and the energy spectrum |X CQ (k)| 2 is calculated; in order to separate the spectral details and the spectral envelope, the CQT energy spectrum is logarithmically obtained log|X CQ (k)| 2; Since the energy spectrum required when performing Discrete Cosine Transform (DCT) is linearly distributed, the log energy spectrum in the geometric distribution state is converted to linear distribution through uniform resampling; and finally the feature extraction of CQCC is completed through DCT transform (the first 40 coefficients are reserved to form a 40-dimensional feature), that is, the CQCC features of the piston rod acceleration signal, the vibration signal and the acoustic emission signal are obtained. The time-frequency domain features and the CQCC features of the piston rod displacement signal, the vibration signal and the acoustic emission signal form the feature vector set of the piston rod displacement signal, the vibration signal and the acoustic emission signal respectively. Similarly, the time-frequency domain features and the CQCC features in the feature vector set are marked with corresponding fault types to obtain fault type labels.

[0083] In an embodiment of the present application, the data signal includes a flow signal and a temperature signal, and feature extraction and fault type marking are performed on the data signal to obtain a feature vector set and a fault type label, including:

[0084] The flow signal and the temperature signal are marked with corresponding fault types to obtain fault type labels.

[0085] The flow signal and the temperature signal are explicit data features, so feature extraction is not required, and corresponding fault type marking can be directly performed by setting the outlet flow of the hydraulic cylinder, the inlet and outlet temperature thresholds to obtain fault type labels.

[0086] Step S300: inputting the feature vector set and the fault type label into a to-be-trained model for model training to obtain a fault diagnosis model.

[0087] The to-be-trained model provided in the embodiments of the present application is an improved deep belief neural network (DBN), and the deep belief neural network has self-adaptive learning ability for data, so that the feature vector set does not need to be manually screened, but can be directly input as a data sample into the to-be-trained model.

[0088] Figure 2 Fig. 2 schematically shows an iterative training flowchart of a training method of a hydraulic cylinder fault diagnosis model according to an embodiment of the present application, Figure 4 Fig. 3 schematically shows a structure of a deep belief neural network according to an embodiment of the present application, which is referred to in conjunction with Figure 2 and Figure 4 In an embodiment of the present application, the to-be-trained model includes a pre-trained model and a top-level classifier, and the feature vector set and the fault type label are input into the to-be-trained model for model training to obtain a fault diagnosis model, including:

[0089] Step S310: inputting the feature vector set into a pre-training model composed of multiple layers of restricted Boltzmann machines for pre-training to obtain a second feature vector set;

[0090] Step S320: inputting the second feature vector set and the fault type label into the top layer classifier for iterative training until a preset iterative training completion condition is met;

[0091] Step S330: determining whether the classification accuracy of the trained model meets a preset first accuracy requirement and whether the error curve of the trained model converges according to a classification result output by the top layer classifier after the iterative training is completed;

[0092] Step S340: in the case where the classification accuracy meets the first accuracy requirement and the error curve converges, the trained model is trained to obtain a fault diagnosis model;

[0093] In the case where the classification accuracy does not meet the first accuracy requirement or the error curve does not converge, the type of the top layer classifier is replaced, and steps S320 and S330 are repeated.

[0094] DBN is a deep neural network composed of RBM (Restricted Boltzmann Machines). DBN mainly includes two steps in the process of training the model. The first step is to train each layer of RBM network in sequence to ensure that as much feature information as possible is retained when the feature vector is mapped to different feature spaces, that is, unsupervised pre-training. The second step is to set a classifier model at the last layer of DBN. Different types of classifier models can be selected according to needs, and the output feature vector of the last RBM is used as the input feature vector of the classifier. The entity relationship classifier is trained in a supervised manner. Then, the error information is propagated from top to bottom to each layer of neurons and each layer of RBM in the network through back propagation, and the entire DBN network is fine-tuned for global optimization. Therefore, the trained model in the embodiment of the application includes a pre-training model (composed of multiple layers of RBM) and a top layer classifier. The first step of model training is to input the feature vector set into the pre-training model composed of multiple layers of restricted Boltzmann machines for pre-training to obtain a second feature vector set.

[0095] In an embodiment of the application, step 310 includes:

[0096] inputting the feature vector set into the pre-training model;

[0097] determining whether the error between the reconstructed vector set output by the pre-training model and the feature vector set meets a preset second accuracy requirement;

[0098] In a case where the error of the reconstruction vector set and the feature vector set does not satisfy the preset second accuracy requirement, the preset learning parameters of the pre-training model are adjusted, and pre-training is performed again.

[0099] In a case where the error of the reconstruction vector set and the feature vector set satisfies the preset second accuracy requirement, the reconstruction vector set is output as the second feature vector set.

[0100] The training effect of the pre-training model needs to be evaluated, and the feature vector output by the pre-training model can be used as the input of the top-level classifier. In a case where the error of the reconstruction feature vector set output by the pre-training model and the input feature vector set satisfies the preset second accuracy requirement after the pre-training is completed, the reconstruction vector set can be output as the second feature vector set. If the error of the reconstruction vector set and the feature vector set does not satisfy the preset second accuracy requirement, the preset learning parameters of the pre-training model need to be adjusted, and the pre-training is performed again until the error of the reconstruction feature vector set and the feature vector set satisfies the preset second accuracy requirement. Specifically, the second accuracy requirement can be adjusted and set according to the accuracy requirement of model training, which is not limited in the present application.

[0101] The second step of model training is to input the second feature vector set output by the pre-training model and the fault type label into the top-level classifier for iterative training, that is, to perform supervised training by using the fault label and the loss function.

[0102] In an embodiment of the present application, the top-level classifier further comprises an endocrine gland for secreting a hormone, and the hormone is used to adjust the input of neurons in the to-be-trained model. Step S320 comprises:

[0103] The second feature vector set and the fault type label are input into the top-level classifier.

[0104] An error value is determined according to a preset error function, a classification result output by the top-level classifier, and the fault type label.

[0105] The concentration of the hormone is adjusted according to a preset concentration decay function and the error value.

[0106] The weights and skewness values of the input of each layer of neurons in the top-level classifier and the weights and skewness values of the input of each layer of neurons in the pre-training model are adjusted according to the error value until the iterative training completion condition is satisfied.

[0107] The endocrine hormone control is a feedback regulation mechanism for maintaining the stability of the internal environment, and the hormone signal can be used to gradually activate and inhibit neurons. An endocrine gland similar to the human feedback regulation mechanism is added to the top-level classifier, and the neurons are adjusted by the hormone concentration, which makes the artificial hormone gland gradually promote or inhibit the output of the neurons according to the data expectation value, so that the neural network is more stable and reliable in training.

[0108] The input of a traditional neuron is the output of the upper layer multiplied by the weight of the current layer:

[0109]

[0110] where X i is the output of the i-th neuron of the upper layer, W i，j is the weight of the current layer, m is the number of neurons of the current layer, n is the number of neurons of the upper layer, and j is the number of the neuron of the current layer, i.e., the j-th neuron of the current layer.

[0111] In the human body, there are multiple different endocrine glands g, which secrete hormones of different concentrations C and act on a certain target cell or several target cells. The target cells have different sensitivities S to different hormones (the sensitivity S can be debugged and set based on training needs), and the hormone release rate r g In the DBN network, there is a similar structure unit to the target cell, which is the neuron. The activation function contained in the neuron does not have error adjustment capability, so the embodiment of the present application introduces endocrine glands to adjust the error value obtained based on the error function calculation of the input data of the global neuron, i.e., to adjust the weight and bias value of the input of each layer of neurons.

[0112] In an embodiment of the present application, the error function is:

[0113]

[0114] where e is the error value, yn(k) is the fault label (i.e., the expected label), y(k) is the output result of the network to be trained, and k is the number of output results of the network to be trained.

[0115] Based on the traditional calculation method, an artificial hormone feedback regulation based on the endocrine mechanism is introduced:

[0116]

[0117] where X i is the output of the upper layer, W i，j is the weight of the current layer, n is the number of neurons of the upper layer, r g is the hormone release rate, S ij is the hormone sensitivity of the current layer, m is the number of neurons of the current layer, and j is the number of the neuron of the current layer, i.e., the j-th neuron of the current layer.

[0118] In human biological system, endocrine glands are controlled by central nervous system to change hormone secretion rate. However, DBN network does not have such regulation, so a hormone concentration decay function must be set, and the control information of the concentration decay function depends on the error function set by the top classifier, that is, if the error value calculated by the error function increases during the training process, the hormone concentration is actively adjusted by reducing the secretion rate, and vice versa. In the network, the hormone concentration secreted by the endocrine glands changes with each iteration of the network, and the concentration is corrected by the error value in each iteration.

[0119] In an embodiment of the present application, the hormone concentration is corrected and adjusted by using the gradient descent method, so the concentration decay function is:

[0120]

[0121] Where C i is the hormone concentration parameter, t is the number of iteration training, e is the error value, and β is the preset learning rate.

[0122] In order to simplify the model operation amount and parameter space, only one set of endocrine glands is used in this embodiment, and the hormone secreted by the endocrine glands and the neuron hormone sensitivity jointly control the neuron input data:

[0123]

[0124] Where X i is the output of the i-th neuron in the upper layer, W i，j is the weight of the current layer, n is the number of neurons in the upper layer, j is the number of neurons in the current layer, that is, the j-th neuron in the current layer, S j is the hormone sensitivity of the current layer, C j is the hormone concentration of the current layer, and m is the number of neurons in the current layer.

[0125] During network training, hormone concentration C may become 0, so that the network will lose some neurons. In order to prevent this from happening, C is multiplied by S and then added to 1, which limits its value to fluctuate between 0 and 1, increasing the rationality of the network. Therefore, the input of each neuron in the trained network can be finally determined as:

[0126]

[0127] Where X i is the output of the i-th neuron in the upper layer, W i，j is the weight of the current layer, n is the number of neurons in the upper layer, j is the number of neurons in the current layer, that is, the j-th neuron in the current layer, S j is the hormone sensitivity of the current layer, and Sj is the hormone sensitivity of the current layer, C j is the hormone concentration of the current layer, b j is the skew value of the current layer, and m is the number of neurons of the current layer.

[0128] In an embodiment of the present application, the gradient descent method is used to correct and adjust the hormone concentration, so the adjustment function of the weight value and the skew value of each layer of neurons is:

[0129]

[0130]

[0131] wherein, W i，j is the weight value, b i is the skew value, t is the number of iterations of the training, e is the error value, and β is the preset learning rate.

[0132] After the iteration training of the to-be-trained model satisfies the preset iteration training completion condition, the processor determines whether the classification accuracy of the to-be-trained model satisfies the preset first accuracy requirement and whether the error curve of the to-be-trained model converges according to the output classification result. In the case where the classification accuracy satisfies the first accuracy requirement and the error curve converges, the training of the to-be-trained model is completed, and a fault diagnosis model is obtained. Specifically, the first accuracy requirement can be debugged and set according to the accuracy requirement of model training, and the present application does not limit this. The iteration training completion condition can be determined based on the number of iterations of the iteration training, or can be determined based on the size of the error value, and the present application does not limit this. In the case where the classification accuracy does not satisfy the first accuracy requirement or the error curve does not converge, the type of the top classifier is replaced, and the second feature vector set is input for iteration training.

[0133] Figure 3 The flowchart of the hydraulic cylinder fault diagnosis method according to an embodiment of the present application is schematically shown, as Figure 3 shown, in an embodiment of the present application, a hydraulic cylinder fault diagnosis method is provided, characterized in that it comprises:

[0134] Step S400: obtaining fault data, wherein the fault data comprises data signals of a plurality of sensors;

[0135] Step S500: performing feature extraction on the fault data to obtain a fault feature vector;

[0136] Step S600: inputting the fault feature vector into a fault diagnosis model to obtain a fault category corresponding to the fault data, wherein the fault diagnosis model is obtained by the training method of the hydraulic cylinder fault diagnosis model as described above.

[0137] The trained hydraulic cylinder fault diagnosis model can be applied to fault diagnosis of the hydraulic cylinder. Fault data of the hydraulic cylinder collected by the plurality of sensors are acquired, and feature extraction is performed on the fault data in the same manner as in the above embodiment to obtain a fault feature vector. The fault feature vector is input into the hydraulic cylinder fault diagnosis model, and the model can automatically classify faults according to the fault feature vector to output a specific fault type.

[0138] According to the above embodiment, a plurality of sensors are used to collect a plurality of physical source data signals representing the state of the hydraulic cylinder. Feature extraction and fault type labeling are performed on the data signals. The extracted feature vectors are used to train a fault diagnosis model. In the subsequent fault diagnosis process, only real-time fault data collected by the sensors need to be input into the trained fault diagnosis model, so that the coupling faults and single faults that may occur simultaneously in the hydraulic cylinder can be classified.

[0139] In an embodiment of the present application, a training device of a hydraulic cylinder fault diagnosis model is provided, comprising:

[0140] A data acquisition unit is configured to acquire data signals of a plurality of sensors, wherein the plurality of sensors are configured to collect a plurality of state data of the hydraulic cylinder.

[0141] A feature extraction and labeling unit is configured to perform feature extraction and fault type labeling on the data signals to obtain a feature vector set and a fault type label, wherein the fault type label comprises a corresponding relationship between a single fault and a related feature vector, and a corresponding relationship between a coupling fault and a related feature vector.

[0142] A model training unit is configured to input the feature vector set and the fault type label into a model to be trained to perform model training, so as to obtain a fault diagnosis model.

[0143] In an embodiment of the present application, a hydraulic cylinder fault diagnosis device is provided, comprising:

[0144] A fault detection unit is configured to acquire fault data, wherein the fault data comprises data signals of a plurality of sensors.

[0145] A fault feature extraction unit is configured to perform feature extraction on the fault data to obtain a fault feature vector.

[0146] A fault classification unit is configured to input the fault feature vector into a fault diagnosis model to obtain a specific fault category, wherein the fault diagnosis model is obtained by using the training method of the hydraulic cylinder fault diagnosis model in the above embodiment.

[0147] In an embodiment of the present application, an electronic device is provided, comprising a processor and a memory, the memory storing machine executable instructions capable of being executed by the processor, the processor executable machine executable instructions implementing the training method of the hydraulic cylinder fault diagnosis model in the above-mentioned embodiments, or the hydraulic cylinder fault diagnosis method in the above-mentioned embodiments.

[0148] In an embodiment of the present application, a machine readable storage medium is provided, the machine readable storage medium storing instructions, the instructions being executed by a processor to implement the training method of the hydraulic cylinder fault diagnosis model in the above-mentioned embodiments, or the hydraulic cylinder fault diagnosis method in the above-mentioned embodiments.

[0149] In an embodiment of the present application, a computer program product is also provided, comprising a computer program, the computer program being executed by a processor to implement the training method of the hydraulic cylinder fault diagnosis model in the above-mentioned embodiments, or the hydraulic cylinder fault diagnosis method in the above-mentioned embodiments.

[0150] Those skilled in the art will understand that embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) having computer usable program code embodied therein.

[0151] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and combinations of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions, which are executed via the processor of the computer or other programmable data processing apparatus, generate a means for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in a flow or multiple flows and / or blocks Figure 1 The functions specified in a flow or multiple flows and / or blocks These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to work in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including instruction means, which implement the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in a flow or multiple flows and / or blocks Figure 1The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks.

[0152] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0153] The memory can include non-persistent memory and / or persistent memory, such as flash memory, read-only memory (ROM), and / or other types of nonvolatile memory. The memory is an example of computer-readable media.

[0154] Computer-readable media includes permanent and non-permanent, removable and non-removable media implemented in any method or technology for storage of information such as computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile discs (DVDs) or other optical storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer-readable media does not include transitory media such as modulated data signals and carrier waves.

[0155] It should also be noted that the terms "comprising," "including," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements recited, but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without more limitations, an element defined by the statement "comprising a" does not exclude the existence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0156] The above merely provides an example of the present application, and is not intended to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall fall into the scope of claims of the present application.

Claims

1. A method of training a hydraulic cylinder fault diagnosis model, characterized by, The method comprises the following steps: acquiring data signals of a plurality of sensors, wherein the plurality of sensors are used to collect a plurality of types of state data of a hydraulic cylinder; performing feature extraction and fault type labeling on the data signals to obtain a feature vector set and a fault type label, wherein the fault type label comprises a corresponding relationship between a single fault and a related feature vector, and a corresponding relationship between a coupling fault and a related feature vector; inputting the feature vector set and the fault type label into a to-be-trained model to perform model training to obtain a fault diagnosis model; wherein the data signals comprise pressure signals, and the performing feature extraction and fault type labeling on the data signals to obtain the feature vector set and the fault type label comprises: performing noise reduction and empirical mode decomposition on the pressure signals to obtain a plurality of intrinsic mode functions and a residual; determining an intrinsic mode energy value, an intrinsic mode energy entropy, a dimensional parameter and a dimensionless parameter of each of the intrinsic mode functions to obtain time domain features of the pressure signals; performing Hilbert transform on each of the intrinsic mode functions, and determining a mean value, a root mean square and a root mean square of a transient frequency of a transient amplitude of each of the transformed intrinsic mode functions to obtain time-frequency domain features of the pressure signals; obtaining a feature vector set of the pressure signals by using the time domain features and the time-frequency domain features of the pressure signals; and obtaining the fault type label by corresponding fault type labeling on the obtained time domain features and time-frequency domain features.

2. The method of claim 1, wherein, The to-be-trained model comprises a pre-trained model and a top-level classifier, and the inputting the feature vector set and the fault type label into the to-be-trained model to perform model training to obtain the fault diagnosis model comprises: step S310: inputting the feature vector set into the pre-trained model composed of a plurality of layers of restricted Boltzmann machines to perform pre-training to obtain a second feature vector set; step S320: inputting the second feature vector set and the fault type label into the top-level classifier to perform iterative training until a preset iterative training completion condition is met; step S330: determining whether the classification accuracy of the to-be-trained model meets a preset first accuracy requirement and whether an error curve of the to-be-trained model converges according to a classification result output by the top-level classifier after the iterative training is completed; step S340: in the case that the classification accuracy meets the first accuracy requirement and the error curve converges, the to-be-trained model training is completed, and the fault diagnosis model is obtained; in the case that the classification accuracy does not meet the first accuracy requirement or the error curve does not converge, replacing the type of the top-level classifier, and repeating steps S320 and S330.

3. The method of claim 2, wherein, The step 310 comprises: inputting the feature vector set into the pre-trained model; determining whether an error between a reconstructed vector set output by the pre-trained model and the feature vector set meets a preset second accuracy requirement; in the case that the error between the reconstructed vector set and the feature vector set does not meet the preset second accuracy requirement, adjusting a preset learning parameter of the pre-trained model to perform pre-training again; In a case where an error between the reconstructed vector set and the feature vector set satisfies a preset second accuracy requirement, output the reconstructed vector set as the second feature vector set.

4. The method of claim 2, wherein, The top-level classifier further comprises endocrine glands for secreting hormones for adjusting inputs of neurons in the to-be-trained model, and the step S320 comprises: inputting the second feature vector set and the fault type label into the top-level classifier; determining an error value according to a preset error function, a classification result output by the top-level classifier, and the fault type label; adjusting a concentration of the hormones according to a preset concentration decay function and the error value; adjusting weights and skewness values of neuron inputs in each layer of the top-level classifier and weights and skewness values of neuron inputs in each layer of the pre-trained model according to the error value until the iteration training completion condition is satisfied.

5. The method of claim 1, wherein, Alternatively, the data signals comprise a piston rod displacement signal, a vibration signal, and an acoustic signal, and the feature extraction and fault type labeling of the data signals to obtain a feature vector set and a fault type label comprise: converting the piston rod displacement signal into a piston rod acceleration signal; frame dividing the piston rod acceleration signal, the vibration signal, and the acoustic signal, and calculating a mean value, a root mean square value, a skewness value, a peak factor, a pulse factor, a waveform factor, a margin factor, and a kurtosis factor of each frame signal to obtain time-frequency domain features of the piston rod acceleration signal, the vibration signal, and the acoustic signal respectively; performing CQCC feature extraction on each frame signal; obtaining a feature vector set of the piston rod displacement signal, the vibration signal, and the acoustic signal respectively by using the time-frequency domain features and the CQCC features of the piston rod displacement signal, the vibration signal, and the acoustic signal respectively; performing corresponding fault type labeling on the obtained time-frequency domain features and CQCC features to obtain the fault type label.

6. A hydraulic cylinder failure diagnosis method characterized by, The method comprises: acquiring fault data, wherein the fault data comprises data signals of multiple sensors; performing feature extraction on the fault data to obtain fault feature vectors; inputting the fault feature vectors into a fault diagnosis model to obtain a fault category corresponding to the fault data, wherein the fault diagnosis model is obtained by the training method of the hydraulic cylinder fault diagnosis model according to any one of claims 1-5.

7. A device for training a hydraulic cylinder fault diagnosis model, characterized by, The method comprises: a data acquisition unit configured to acquire data signals of multiple sensors, wherein the multiple sensors are configured to collect multiple types of state data of a hydraulic cylinder; a feature extraction and labeling unit configured to perform feature extraction and fault type labeling on the data signals to obtain a feature vector set and a fault type label, wherein the fault type label comprises a corresponding relationship between a single fault and a related feature vector, and a corresponding relationship between a coupled fault and a related feature vector. The model training unit is configured to input the feature vector set and the fault type label into a model to be trained to perform model training, so as to obtain a fault diagnosis model. The feature extraction and fault type labeling of the data signal to obtain the feature vector set and the fault type label include: performing noise reduction and empirical mode decomposition on the pressure signal to obtain a plurality of intrinsic mode functions and a residual; determining the intrinsic mode energy value, the intrinsic mode energy entropy, the dimensional parameter and the dimensionless parameter of each intrinsic mode function to obtain the time domain feature of the pressure signal; performing Hilbert transform on each intrinsic mode function, and determining the mean value, the root mean square and the root mean square of the transient frequency of the transient amplitude of each transformed intrinsic mode function to obtain the time-frequency domain feature of the pressure signal; obtaining the feature vector set of the pressure signal by using the time domain feature and the time-frequency domain feature of the pressure signal; and performing corresponding fault type labeling on the obtained time domain feature and time-frequency domain feature to obtain the fault type label.

8. A hydraulic cylinder failure diagnosis device characterized by comprising: The method comprises: The fault detection unit is configured to obtain fault data, wherein the fault data comprises data signals of a plurality of sensors; The fault feature extraction unit is configured to perform feature extraction on the fault data to obtain a fault feature vector; The fault classification unit is configured to input the fault feature vector into a fault diagnosis model to obtain a specific fault category, wherein the fault diagnosis model is obtained by using the hydraulic cylinder fault diagnosis model training method according to any one of claims 1 to 5.

9. An electronic device, comprising: The processor and the memory are included, and the memory stores machine executable instructions which can be executed by the processor. The processor can execute the machine executable instructions to implement the hydraulic cylinder fault diagnosis model training method according to any one of claims 1 to 5, or the hydraulic cylinder fault diagnosis method according to claim 6.

10. A machine-readable storage medium, characterized in that, The machine readable storage medium stores instructions which, when executed by a processor, cause the processor to implement the hydraulic cylinder fault diagnosis model training method according to any one of claims 1 to 5, or the hydraulic cylinder fault diagnosis method according to claim 6.

Images