Precision speed reducer intelligent fault diagnosis method based on multi-dimensional encoder signals

Abstract

The invention relates to the technical field of speed reducer fault diagnosis, in particular to a precise speed reducer intelligent fault diagnosis method based on multi-dimensional encoder signals. The method comprises the steps of collecting an encoder signal of an input end of a reducer, and extracting an input angle position of the input end; collecting an encoder signal at the output end of the speed reducer, and extracting an output angle position of the output end; obtaining the transmission ratio of the reducer, and calculating a transmission error according to the input angular position, the output angular position and the transmission ratio; calculating an output instantaneous angular velocity and an output instantaneous angular acceleration corresponding to the output angular position; performing data preprocessing on the transmission error, the output instantaneous angular velocity and the output instantaneous angular acceleration to obtain multi-dimensional fusion data; and inputting the multi-dimensional fusion data into a pre-trained fault diagnosis model to obtain the running state of the speed reducer. Reducer fault feature recognition is carried out based on the fused multidimensional encoder time domain signals, the problem that the signal type for fault diagnosis is single is solved, and the fault diagnosis precision is improved.

Description

Technical Field

[0001] The invention relates to the technical field of speed reducer fault diagnosis, and in particular to an intelligent fault diagnosis method for a precision speed reducer based on multi-dimensional encoder signals. Background Art

[0002] As a key core component of industrial mother machines, industrial robots and other equipment, precision reducers are generally under high load and high torque working conditions, prone to failure or performance degradation, causing production stagnation. Therefore, it is of great significance to perform fault detection on precision reducers.

[0003] The patent application document with application number CN202211680843.1 discloses a transmission system gearbox fault diagnosis method based on the encoder transmission error signal. The transmission error signal is calculated by collecting the input shaft and output shaft encoder signals. After the transmission error signal is detrended and zero-mean processed, the speed fluctuation trend term is extracted by variational mode decomposition, and the speed fluctuation is decoupled to obtain the detrended transmission error signal. Finally, through time domain synchronous average analysis and FFT spectrum analysis, it is determined whether the gearbox is faulty based on the changes in the meshing frequency and sideband amplitude under the fault state. However, fault diagnosis is performed only through the transmission error signal, and the signal type is single. With the normal use of the equipment, there is a problem of poor fault diagnosis effect due to signal interference caused by wear.

[0004] Therefore, the present application proposes a precision reducer intelligent fault diagnosis method based on multi-dimensional encoder signals to perform feature recognition on reducer faults to solve the problem of a single signal type used for diagnosis and improve the accuracy of fault diagnosis. Summary of the invention

[0005] The purpose of the present invention is to address the technical problems that the signal type used for fault diagnosis in the prior art is single and the fault diagnosis effect of the precision reducer is not good. The intelligent fault diagnosis method for precision reducers based on multi-dimensional encoder signals provided by the embodiments of the present application adopts dual encoders at the input and output ends, thereby avoiding the problem of difficult installation of external encoders and the hidden danger of introducing additional fault sources caused by the use of multiple encoders. The instantaneous angular velocity, instantaneous angular acceleration and transmission error data are obtained through signal processing, and the fault characteristics of the reducer are identified using the instantaneous angular velocity, instantaneous angular acceleration and transmission error data, solving the problem of single signal type. At the same time, a fault diagnosis model for multi-dimensional encoder signal input is constructed to realize the intelligent and automated extraction, monitoring and diagnosis of fault characteristics.

[0006] In order to achieve the above-mentioned object of the invention, the present invention provides the following technical solutions: Intelligent fault diagnosis method for precision reducer based on multi-dimensional encoder signal, including Collecting the encoder signal of the reducer input end and extracting the input angular position of the input end; collecting the encoder signal of the reducer output end and extracting the output angular position of the output end; Obtaining the transmission ratio of the reducer, calculating the transmission error according to the input angular position, the output angular position and the transmission ratio; calculating the output instantaneous angular velocity and the output instantaneous angular acceleration corresponding to the output angular position; The transmission error, the output instantaneous angular velocity and the output instantaneous angular acceleration are subjected to data preprocessing to obtain multi-dimensional fusion data; the multi-dimensional fusion data is input into a pre-trained fault diagnosis model to obtain the operating state of the reducer.

[0007] As a preferred technical solution of the present application, the method for calculating the transmission error includes: ; in, is the transmission error; The output angular position collected at the output end at time t; is the input angular position collected at the input end at time t; i is the transmission ratio.

[0008] As a preferred technical solution of the present application, the method for calculating the output instantaneous angular velocity includes: ; in, is the output instantaneous angular velocity; For the time interval.

[0009] As a preferred technical solution of the present application, the method for calculating the output instantaneous angular acceleration includes: ; in, is the output instantaneous angular acceleration.

[0010] As a preferred technical solution of the present application, the method of performing data preprocessing on the transmission error, the output instantaneous angular velocity and the output instantaneous angular acceleration includes: The transmission error, the output instantaneous angular velocity and the output instantaneous angular acceleration are normalized respectively; a first number and a sliding window are preset, the corresponding normalized results are intercepted according to the first number, and the intercepted data slices are spliced ​​according to a preset splicing order to obtain the multi-dimensional fusion data; the sliding window is moved to intercept the next data slice for data processing; the first number is greater than the periodic sampling number of the encoder.

[0011] As a preferred technical solution of the present application, the calculation formula for the normalization processing includes: ; in, is the transmission error or the output instantaneous angular velocity or the output instantaneous angular acceleration; is the corresponding normalized result.

[0012] As a preferred technical solution of the present application, the fault diagnosis model receives the multidimensional fusion data as input, extracts waveform features from the multidimensional fusion data; extracts autocorrelation features based on the periodic signal data of the data slices in a single dimension; extracts cross-correlation features based on the similarity of the data slices in different dimensions, performs diagnosis based on the waveform features, the autocorrelation features and the cross-correlation features, and outputs the operating status of the encoder.

[0013] As a preferred technical solution of the present application, the fault diagnosis model is established based on a residual network, including: an input layer, a first residual network, a second residual network and an output layer; The first residual network includes a first residual block and a first pooling layer, and the first residual block includes a first convolutional layer, a first batch normalization layer, a second convolutional layer, and a second batch normalization layer; The second residual network includes a second residual block and a second pooling layer, and the second residual block includes a third convolutional layer, a third batch normalization layer, a fourth convolutional layer and a fourth batch normalization layer.

[0014] As a preferred technical solution of the present application, the sizes of the first convolutional layer, the second convolutional layer and the first pooling layer are equal; The third convolutional layer, the fourth convolutional layer and the second pooling layer are equal in size.

[0015] As a preferred technical solution of the present application, the sizes of the first convolutional layer, the second convolutional layer and the first pooling layer are recorded as a first magnitude; The sizes of the third convolutional layer, the fourth convolutional layer, and the second pooling layer are recorded as a second magnitude; The first magnitude is the second magnitude times.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. The fault diagnosis method proposed in the present invention is based on the encoder signals at the input and output ends, and has the characteristics of short transmission path and higher signal fidelity. The original encoders at the input and output ends are used to extract the angular position signals, which reduces the difficulty of installing external encoders and reduces additional fault sources. Based on the angular position signal obtained by the encoder, time domain signals of multiple dimensions are obtained through signal conversion, including: transmission error, output instantaneous angular velocity and output instantaneous angular acceleration. A fault diagnosis model is constructed to identify the reducer fault characteristics of the fused multi-dimensional encoder time domain signal, which solves the problem of single signal type, improves the fault diagnosis accuracy, and realizes a precision reducer intelligent fault diagnosis method based on multi-dimensional encoder signals.

[0017] 2. The transmission error is calculated by the difference between the encoder signals at the input and output ends, which reflects the transmission accuracy inside the reducer and reduces the interference of the complex structure inside the reducer on fault diagnosis. The output instantaneous angular velocity and output instantaneous angular acceleration are obtained by differential calculation of the encoder signal at the output end, reflecting the dynamic characteristics of the reducer. These three time domain signals are calculated based on the angular position signal, and any fault in the rotating machinery will directly affect the angular change of the rotating shaft. Therefore, the encoder time domain signal contains rich information that can reflect the health status and operating conditions of the equipment. By processing the angular position signal in multiple dimensions to obtain time domain signals, the limitations of fault diagnosis based on a single signal are reduced. The encoder signal data can be further used to extract fault features to more comprehensively reflect the health status of the reducer.

[0018] 3. The first number of intercepted data slices is greater than the number of periodic sampling of the encoder, so that the data slices have periodic time domain signal characteristics. Time domain signals of different dimensions are calculated based on the angular position signals at the input and output ends, and there are similarities. The fault diagnosis model extracts the waveform characteristics, periodic autocorrelation characteristics and similar cross-correlation characteristics of the multidimensional time domain signal through the residual network. These features can help the fault diagnosis model understand the basic form of the time domain signal, automatically identify the mutation points, jumps or slow trend changes in the time domain signal, improve the accuracy of the model in diagnosing faults, and realize the intelligent diagnosis of the reducer operating status. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a schematic diagram of the structure of the worm gear precision reducer of the test device of this application; Figure 2 A flowchart of the fault diagnosis method of the present application; Figure 3 It is a structural diagram of the fault diagnosis model; Figure 4 Schematic diagram of the structure of the first residual network; Figure 5 Schematic diagram of the structure of the second residual network; Figure 6 is a schematic diagram of the structure of the first residual block; Figure 7 Schematic diagram of the structure of the second residual block; Figure 8 Flowchart for constructing a fault diagnosis model based on multi-dimensional encoder signal fusion; Fig. 9 It is a waveform diagram of splicing multi-dimensional fusion data; Fig.10 Loss function curve graph for model training and validation; Fig.11 Accuracy curves for model training and verification; Fig.12 This is the confusion matrix diagram of the model test results; Markings in the figure: 1-driving motor, 2-input shaft coupling, 3-input end encoder, 4-test reducer, 5-output end encoder, 6-output shaft coupling, 7-load motor. DETAILED DESCRIPTION

[0020] To make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be described clearly and completely in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them.

[0021] Therefore, the following detailed description of the embodiments of the present invention is not intended to limit the scope of the invention claimed for protection, but merely represents some embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features and technical solutions in the embodiments may be combined with each other.

[0023] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0024] In the description of the present invention, it should be noted that the orientation or positional relationship indicated by the terms "upper", "lower", etc. is based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the invention product is usually placed when in use, or the orientation or positional relationship commonly understood by those skilled in the art. Such terms are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0025] The technical solution proposed in the present invention is based on the encoder signals at the input and output ends, and performs fault diagnosis through a pre-trained fault diagnosis model, which is applicable to various types of precision reducers. Figure 1 As shown, the worm gear precision reducer is used as a specific implementation object for detailed explanation.

[0026] The worm gear precision reducer experimental platform in this embodiment includes: a driving motor 1, an input shaft coupling 2, an input end encoder 3, a test reducer 4, an output end encoder 5, an output shaft coupling 6, and a load motor 7. The driving motor 1 is connected to the input end of the worm gear test reducer 4 through the input shaft coupling 2, and is used to provide a driving speed and a driving torque for the worm gear test reducer 4; the load motor 7 is connected to the output end of the worm gear test reducer 4 through the output shaft coupling 6, and is used to provide a load for the worm gear test reducer 4; the input end and the output end of the worm gear test reducer 4 are both arranged with encoders, which are used to obtain the angular position signals of the reducer input end and the output end.

[0027] In one embodiment of the present application, for example Figure 2 As shown, the precision reducer intelligent fault diagnosis method based on multi-dimensional encoder signals provided in this embodiment includes: The encoder data acquisition card is used to collect the encoder signal at the input end of the reducer, and the input angular position of the input end is extracted and recorded as ; Collect the encoder signal of the reducer output end, extract the output angular position of the output end, and record it as .

[0028] Obtain the transmission ratio i of the reducer, according to the input angular position , the output angular position and the transmission ratio i to calculate the transmission error In this embodiment, the method for calculating the transmission error includes: ; in, The output angular position collected at the output end at time t; is the input angular position collected at the input end at time t. The transmission error is calculated based on the encoder signals at the input and output ends, and is used to reflect the transmission accuracy inside the reducer as a whole. For the same system, analyzing the reducer fault through the transmission error signal can reduce the interference of the reducer due to external factors; at the same time, since there is no need to pay attention to the complex structure and complex operating conditions inside the reducer, it is conducive to reducing the analysis factors in the fault analysis process and improving the efficiency of fault analysis.

[0029] Calculate the output instantaneous angular velocity corresponding to the output angular position signal And output instantaneous angular acceleration In this embodiment, the output instantaneous angular velocity is obtained by applying the differential method to the output angular position. , the formula is: ; The output instantaneous angular acceleration is obtained by applying the differential method to the output instantaneous angular velocity. , the formula is: ; Furthermore, the reducer converts the power at the input end into rotational motion at the output end through meshing transmission. Failure of the reducer will affect both ends, but the angular position signal collected at the input end may be interfered by the drive system, while the angular position signal collected at the output end can more directly reflect the signal change caused by the reducer failure, which is conducive to extracting the corresponding fault characteristics. Therefore, in this embodiment, the transmission error, the output instantaneous angular velocity and the output instantaneous angular acceleration are preprocessed to obtain multi-dimensional fusion data; the multi-dimensional fusion data is input into a pre-trained fault diagnosis model to obtain the operating state of the reducer, including the type of normal working condition or fault state of the reducer, such as worm tooth surface wear, worm wheel tooth surface wear, worm wheel and worm center distance error, and worm axial error.

[0030] The fault diagnosis method proposed in the present invention is based on the encoder signals at the input and output ends, and has the characteristics of short transmission path and higher signal fidelity. The original dual encoders at the input and output ends are used to extract signals, which reduces the difficulty of installing external encoders and reduces additional fault sources. Based on the angular position signal obtained by the encoder, time domain signals of multiple dimensions are obtained through signal conversion, including: transmission error, output instantaneous angular velocity and output instantaneous angular acceleration. A fault diagnosis model is constructed to identify the reducer fault characteristics of the fused multi-dimensional encoder time domain signal, solve the problem of single signal type, improve the fault diagnosis accuracy, and realize a precision reducer intelligent fault diagnosis method based on multi-dimensional encoder signals.

[0031] The transmission error is calculated by the difference between the encoder signals at the input and output ends, which reflects the transmission accuracy inside the reducer and reduces the interference of the complex structure inside the reducer on fault diagnosis. The output instantaneous angular velocity and output instantaneous angular acceleration are obtained by differential calculation of the encoder signal at the output end, reflecting the dynamic characteristics of the reducer. These three time domain signals are calculated based on the angular position signal, and any fault in the rotating machinery will directly affect the angular change of the rotating shaft. Therefore, the encoder time domain signal contains rich information that can reflect the health status and operating conditions of the equipment. By processing the angular position signal in multiple dimensions, time domain signals are obtained, which reduces the limitations of fault diagnosis based on a single signal. The encoder signal data can be further used to extract fault features to more comprehensively reflect the health status of the reducer.

[0032] In one embodiment of the present application, the method for performing data preprocessing on the transmission error, the output instantaneous angular velocity and the output instantaneous angular acceleration includes: The transmission error, the output instantaneous angular velocity and the output instantaneous angular acceleration are normalized respectively. A first quantity and a sliding window are preset, and the corresponding normalized results are intercepted according to the first quantity, and the intercepted data slices are spliced ​​according to a preset splicing order to obtain the multi-dimensional fusion data. The sliding window is moved to intercept the next data slice for data processing. The first quantity is greater than the periodic sampling quantity of the encoder. The periodic sampling quantity is the number of sampling points of the encoder in a single cycle, and the single cycle is the time required for the output shaft to rotate one circle.

[0033] Specifically, the transmission error, the output instantaneous angular velocity and the output instantaneous angular acceleration are respectively entered into corresponding CSV files, and all data are normalized by a preset normalization formula so that the mean value of the data is 0 and the standard deviation is 1. In this embodiment, the calculation formula for the normalization process includes: ; in, is the transmission error or the output instantaneous angular velocity or the output instantaneous angular acceleration; is the corresponding normalized result.

[0034] Furthermore, data is read from the normalized CSV file. In this embodiment, the read data can be directly sliced ​​as numerical sequence data, or it can be sliced ​​based on the data time series to form time domain signal waveform image data. After the slicing process, input data for fault diagnosis is obtained. In a specific embodiment, the first number is slightly larger than the periodic sampling number, so that the data slice can not only contain time domain signal data with a periodic waveform, but also control the amount of data. At the same time, in order to make it easier to optimize memory access and data alignment when the hardware device processes data, the first number is set to , m is a positive integer.

[0035] In this embodiment, when the output end of the reducer rotates one circle, the number of sampling points of the encoder is 1700. 2048 sampling points are intercepted from each normalized signal data as a data slice, and each data slice is spliced ​​according to a preset splicing order to obtain the multi-dimensional fusion data. A too short length of the sliding window and a too long length of the data slice will result in a larger amount of calculation and consume more computing resources; a too long length of the sliding window may cause the loss of local information of the signal. Therefore, in this embodiment, the length of the sliding window is at most half the length of the data slice. Specifically, the length of the sliding window is set to 1000 sampling points to ensure that the periodic time domain signal data has more repeated parts to increase the diversity of the sampling point data, which helps to capture the dynamic changes and local features in the encoder signal.

[0036] Furthermore, the preset splicing order is determined during the training process of the fault diagnosis model, and this embodiment does not impose any specific order restriction.

[0037] In one embodiment of the present application, the fault diagnosis model receives the multidimensional fusion data as input, extracts waveform features from the multidimensional fusion data; extracts autocorrelation features based on the periodic signal data of the data slices in a single dimension; extracts cross-correlation features based on the similarity of the data slices in different dimensions, performs diagnosis based on the waveform features, the autocorrelation features and the cross-correlation features, and outputs the operating status of the encoder.

[0038] Specifically, the fault diagnosis model is constructed based on a residual network. The waveform features are extracted through the residual network, including: mean features, which are used to represent the central trend of the signal; standard deviation features, which are used to measure the degree of signal fluctuation; maximum features and / or minimum features, which are used to reflect the dynamic range of the signal; skewness features, which are used to measure the degree of skewness of the signal distribution; and kurtosis features, which are used to measure the steepness of the signal distribution morphology.

[0039] Since the first number is greater than the periodic sampling number, the data slice has a periodic time domain signal feature, and this periodic time domain signal feature reflects the similarity between the time domain signal and its own time delayed version. Therefore, during the training process of the fault diagnosis model, the periodic time domain signal features are extracted through the residual network to obtain the autocorrelation features, which helps to capture the dynamic changes and local features in the encoder signal. After the data slices of different dimensions are spliced, the multi-dimensional fused data is obtained. The fault diagnosis model can extract the autocorrelation features of multiple dimensions, which improves the model's learning intensity of the periodic features of the time domain signal, which is conducive to improving the accuracy of the model in diagnosing faults.

[0040] At the same time, in this embodiment, the time domain signals of different dimensions as the model input are all calculated based on the angular position signal extracted by the encoder. Therefore, there are similarities between the time domain signals of different dimensions. The similarities of the time domain signals of multiple dimensions are extracted by the residual network to obtain the cross-correlation features to capture the mutual dependence between the signals.

[0041] The fault diagnosis model extracts the waveform features, autocorrelation features and cross-correlation features of the multidimensional time domain signal through the residual network. These features can help the fault diagnosis model understand the basic form of the time domain signal, automatically identify the mutation points, jumps or slow trend changes in the time domain signal, and realize intelligent diagnosis of the reducer operation status.

[0042] In one embodiment of the present application, for example Figure 3-Figure 7 As shown, the structure of the fault diagnosis model includes: an input layer, a first residual network, a second residual network and an output layer.

[0043] The first residual network includes a first residual block and a first pooling layer, and the first residual block includes a first convolutional layer, a first batch normalization layer, a second convolutional layer, and a second batch normalization layer; The second residual network includes a second residual block and a second pooling layer, and the second residual block includes a third convolutional layer, a third batch normalization layer, a fourth convolutional layer and a fourth batch normalization layer.

[0044] Among them, the output layer includes a Dropout layer and a flatten layer, and then two fully connected layers are used to classify the operating status of the reducer. A Dropout layer is set between the fully connected layers, and finally a Softmax layer is used to output the classification probability of the diagnosis results.

[0045] Furthermore, the first convolution layer, the second convolution layer and the first pooling layer are of equal size; the third convolution layer, the fourth convolution layer and the second pooling layer are of equal size.

[0046] Furthermore, the sizes of the first convolutional layer, the second convolutional layer and the first pooling layer are recorded as a first magnitude; the sizes of the third convolutional layer, the fourth convolutional layer and the second pooling layer are recorded as a second magnitude; the first magnitude is a multiple of the second magnitude. times, n is a positive integer.

[0047] The training flow chart of the fault diagnosis model is as follows: Figure 8 As shown. Before model training, it is necessary to obtain the encoder signals of the input and output ends under different operating states and enter them into the CSV file of the corresponding operating state, so as to realize the label classification of the encoder signal and the corresponding operating state. In this embodiment, the label classification includes: normal operating conditions, worm tooth surface wear, worm wheel tooth surface wear, worm wheel and worm center distance error, and worm axial error. For other types of fault diagnosis training, it can be implemented according to the training method provided in this embodiment, and will not be listed one by one.

[0048] The transmission error, the output instantaneous angular velocity and the output instantaneous angular acceleration are obtained by calculation based on the angular position signals collected by the encoders at the input and output ends. Fig. 9 As shown, 2048 sampling point data are intercepted for three time domain signals to obtain the data slices. The data slices are spliced ​​head to tail to achieve the fusion of the feature information of the three signals, and the multidimensional fusion data of size 6144*1 is obtained. The data slices are spliced ​​in different orders to obtain multiple first sets corresponding to the splicing order. Multiple first models are constructed based on the residual network for training the first sets with different splicing orders. The first set is input into the first model, and after setting the hyperparameters, optimizer, loss function, etc., the model is trained and verified to obtain the first model with the highest accuracy of fault diagnosis. The first model is used to perform classification test on the test set, and the confusion matrix and accuracy curve are drawn to represent the classification effect of the model. Observe the fault diagnosis indicators of the first model on the test set, such as the accuracy of fault classification, select the first model with the highest accuracy as the fault diagnosis model, and the splicing order corresponding to the model is used as the preset splicing order; otherwise, adjust the hyperparameters of the first model and continue model training.

[0049] Taking the splicing sequence of the transmission error, the output instantaneous angular velocity and the output instantaneous angular acceleration as an example, the first set corresponding to the splicing sequence is divided into a training set, a validation set and a test set in a ratio of 7:2:1. Figure 3As shown, the input layer is the multi-dimensional fusion data obtained after intercepting and splicing the signal data. The fault diagnosis model adjusts the number of channels of the feature through the convolution kernel. The residual block extracts the corresponding time domain signal features through the convolution layer, and the batch normalization layer and the ReLU activation function increase the nonlinearity of the model and improve the generalization ability of the model. In order to avoid the disappearance or explosion of gradients in deep networks, the shortcut layer directly adds the input to the output after convolution processing to form residual learning. The residual connection allows the gradient to be transferred more easily, thereby reducing the difficulty of training deep networks. The batch normalization layer can accelerate training and stabilize the model by normalizing the output after convolution, and the ReLU activation function is used to enhance the nonlinearity of the model. The pooling layer performs maximum pooling on the output of the residual block, shortens the time series length of the data, and keeps the number of channels unchanged. By reducing the size of the input data and maintaining important features, it helps to prevent overfitting and reduce the amount of calculation.

[0050] The Dropout layer is used in the output layer to avoid overfitting, and a part of the nodes of the neural network are randomly discarded. In this embodiment, the dropout rate is set to 50%. The fully connected layer outputs 5 neurons, representing 5 classification tasks. The Softmax layer converts the output of the fully connected layer into a probability distribution, and each output value represents the probability that the sample belongs to a certain category.

[0051] Specifically, the formula for the convolution operation of the convolution layer is: ; in, Represents the output vector y of the lth layer; Indicates The cth channel in the layer; represents the convolution kernel parameter weight matrix of the lth layer, where i represents the i-th channel in the lth layer and c represents the The cth channel of the layer; represents the convolution operation, Indicates Layer The output vector x of the channel is also the input vector of this layer; Indicates The bias term b for the i-th channel of the layer.

[0052] Specifically, the formula of the activation function layer ReLU is: ; in, express The activation value after the activation function; represents the jth weight of the i-th feature vector in convolutional layer l.

[0053] Specifically, the calculation formula of the maximum pooling layer is: ; in, represents the output value of the jth neuron in the ith channel in the lth layer; w represents the width of the pooling kernel; Represents the input value of the tth neuron in the ith channel in the lth layer.

[0054] Specifically, the calculation formula of the Softmax layer classification probability is: ; in, represents the i-th neuron input of the output layer; C represents the number of categories of running status, which is 5 in this embodiment; Represents the classification probability of the corresponding running state.

[0055] In this embodiment, during model training, the maximum number of training times is set to 125, the sample size for each batch of training is set to 8, the initial learning rate is set to 0.0165, a regularization term is used to prevent overfitting, the weight decay factor is set to 0.005, and the SGD optimizer is selected to calculate and update the parameters of the model. The optimizer momentum is 0.9, and an exponential decay learning rate scheduler is used. Each time the learning rate is updated, the learning rate is multiplied by 0.95, so that the learning rate gradually decreases during the training process, which helps to stabilize the convergence in the later stage of training.

[0056] The training effect of the fault diagnosis model is shown in Figure 10-11 The test data is substituted into the model to evaluate the classification effect. The diagnosis results of this embodiment are shown in Fig.12 As shown, judging from the confusion matrix results of the fault diagnosis model, the present invention effectively improves the accuracy of reducer fault diagnosis.

[0057] In one embodiment of the present application, a precision reducer intelligent fault diagnosis device based on a multidimensional encoder signal is also provided, the device comprising: an encoder for collecting reducer input and output signals; at least one processor capable of processing the reducer input and output signals collected by the encoder; at least one memory for storing the reducer input and output signals collected by the encoder and at least one program; when the at least one program is executed by the at least one processor, so that the at least one processor implements the precision reducer intelligent fault diagnosis method based on the multidimensional encoder signal, the processor and the memory can be connected to store at least one program through a bus or other means; the memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory can include a high-speed random access memory, and can also include a non-transient memory, such as at least one disk storage device, a flash memory device or other non-transient solid-state storage device. In some embodiments, the memory optionally includes a memory remotely arranged relative to the processor, and these remote memories can be connected to the device through a network. Examples of the above-mentioned network include but are not limited to the Internet, corporate intranet, local area network, mobile communication network and combinations thereof. When the at least one program is executed by the at least one processor, the at least one processor implements the precision reducer intelligent fault diagnosis method based on multi-dimensional encoder signals.

[0058] In one embodiment of the present application, a computer storage medium is also provided, in which a program executable by a processor is stored. When the program executable by the processor is executed by the processor, it is used to implement the intelligent fault diagnosis method of the precision reducer based on the multi-dimensional encoder signal.

[0059] It will be appreciated by those skilled in the art that all or some of the steps and systems in the disclosed method above may be implemented as software, firmware, hardware and appropriate combinations thereof. Some physical components or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor or a microprocessor, or may be implemented as hardware, or may be implemented as an integrated circuit, such as an application specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or a non-transitory medium) and a communication medium (or a temporary medium). As known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, disk storage or other magnetic storage devices, or any other medium that may be used to store desired information and may be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically embodies computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0060] The above embodiments are only used to illustrate the present invention and are not intended to limit the technical solutions described in the present invention. Although the present invention has been described in detail with reference to the above embodiments, the present invention is not limited to the above specific implementation methods. Therefore, any modification or equivalent replacement of the present invention; and all technical solutions and improvements thereof that do not depart from the spirit and scope of the invention are included in the scope of the claims of the present invention.

Claims

1. An intelligent fault diagnosis method for precision reducers based on multi-dimensional encoder signals, characterized in that: include: Collecting the encoder signal of the reducer input end and extracting the input angular position of the input end; Collecting the encoder signal of the output end of the reducer to extract the output angular position of the output end; Acquire the transmission ratio of the reducer, and calculate the transmission error according to the input angular position, the output angular position and the transmission ratio; Calculating the output instantaneous angular velocity and the output instantaneous angular acceleration corresponding to the output angular position; The transmission error, the output instantaneous angular velocity and the output instantaneous angular acceleration are subjected to data preprocessing to obtain corresponding data slices; multi-dimensional fusion data are obtained according to the data slices; the multi-dimensional fusion data is input into a pre-trained fault diagnosis model to obtain the operating status of the reducer.

2. The intelligent fault diagnosis method for precision reducer based on multi-dimensional encoder signal according to claim 1 is characterized in that: The method for calculating the transmission error includes: ； in, is the transmission error; The output angular position collected at the output end at time t; is the input angular position collected at the input end at time t; i is the transmission ratio.

3. The intelligent fault diagnosis method for precision reducer based on multi-dimensional encoder signal according to claim 1 is characterized in that: The method for calculating the output instantaneous angular velocity comprises: ； in, is the output instantaneous angular velocity; is the time interval.

4. The intelligent fault diagnosis method for precision reducer based on multi-dimensional encoder signal according to claim 3 is characterized in that: The method for calculating the output instantaneous angular acceleration comprises: ； in, is the output instantaneous angular acceleration.

5. The intelligent fault diagnosis method for precision reducer based on multi-dimensional encoder signal according to claim 1, characterized in that: The method of performing data preprocessing on the transmission error, the output instantaneous angular velocity and the output instantaneous angular acceleration comprises: The transmission error, the output instantaneous angular velocity and the output instantaneous angular acceleration are normalized respectively; a first number and a sliding window are preset, the corresponding normalized results are intercepted according to the first number, and the data slices obtained after the interception are spliced ​​according to a preset splicing order to obtain the multi-dimensional fusion data; the sliding window is moved to intercept the next data slice for data processing; the first number is greater than the periodic sampling number of the encoder.

6. The intelligent fault diagnosis method for precision reducer based on multi-dimensional encoder signal according to claim 5 is characterized in that: The calculation formula for the normalization process includes: ； in, is the transmission error or the output instantaneous angular velocity or the output instantaneous angular acceleration; is the corresponding normalized result.

7. The intelligent fault diagnosis method for precision reducer based on multi-dimensional encoder signal according to claim 1, characterized in that: The fault diagnosis model receives the multi-dimensional fusion data as input and extracts waveform features from the multi-dimensional fusion data; Autocorrelation features are extracted from the periodic signal data of the data slices in a single dimension; cross-correlation features are extracted based on the similarity of the data slices in different dimensions, diagnosis is performed based on the waveform features, the autocorrelation features and the cross-correlation features, and the operating status of the encoder is output.

8. The intelligent fault diagnosis method for precision reducer based on multi-dimensional encoder signal according to claim 1, characterized in that: The fault diagnosis model is established based on a residual network, and includes: an input layer, a first residual network, a second residual network and an output layer; The first residual network includes a first residual block and a first pooling layer, and the first residual block includes a first convolutional layer, a first batch normalization layer, a second convolutional layer, and a second batch normalization layer; The second residual network includes a second residual block and a second pooling layer, and the second residual block includes a third convolutional layer, a third batch normalization layer, a fourth convolutional layer and a fourth batch normalization layer.

9. The intelligent fault diagnosis method for precision reducer based on multi-dimensional encoder signal according to claim 8, characterized in that: The first convolutional layer, the second convolutional layer, and the first pooling layer are of equal size; The third convolutional layer, the fourth convolutional layer and the second pooling layer are equal in size.

10. The intelligent fault diagnosis method for precision reducer based on multi-dimensional encoder signal according to claim 8, characterized in that: The sizes of the first convolutional layer, the second convolutional layer, and the first pooling layer are recorded as a first magnitude; The sizes of the third convolutional layer, the fourth convolutional layer, and the second pooling layer are recorded as a second magnitude; The first magnitude is the second magnitude times.

Images