A fault diagnosis method and system for crane hook lifting motor

Abstract

The present invention provides a fault diagnosis method and system for a crane hook lifting motor, which relates to the field of data processing technology. The method includes: obtaining a vibration signal, a current signal and an audio signal of the crane hook lifting motor; using a normalized vibration feature map as an R channel, a normalized current feature map as a G channel, and a normalized audio feature map as a B channel to generate a color feature map; constructing a convolutional neural network with a cascade structure; inputting the color feature map to the input layer of the convolutional neural network; and performing fault diagnosis through the convolutional neural network. The present invention can map the vibration signal, the current signal and the audio signal into two-dimensional feature maps respectively, and fuse them into a color feature map through the RGB three channels, so that the convolutional neural network can more easily extract useful information from these feature maps, and improve the accuracy of the convolutional neural network for the fault diagnosis of the lifting motor.

Description

Technical Field

[0001] The invention relates to the technical field of data processing, and in particular to a fault diagnosis method and system for a crane hook lifting motor. Background Art

[0002] A crane is a large mechanical device used to transport materials vertically and horizontally. It can lift, lower or move heavy objects. The hook lift motor is a key component of the crane and is responsible for controlling the lifting and lowering of the hook. The hook is a device used by the crane to grab and move materials. The hook lift motor controls the winch to lift or lower the hook to move the heavy objects. Its function is to convert electrical energy into mechanical energy for lifting the load. Once the hook lift motor fails, it may seriously affect the normal operation of the crane and even cause equipment damage or safety accidents.

[0003] At present, with the rapid development of science and technology, more and more modern technologies are being applied to motor fault diagnosis. The typical detection technology is to use deep learning technology to perform fault diagnosis based on the vibration signal of the motor.

[0004] However, the current deep learning fault diagnosis technology based on vibration signals has low diagnostic accuracy during the diagnosis process. The main reason is that when only vibration signals are used for fault diagnosis, many fault characteristics of the motor may not be fully reflected in the vibration. For example, some electrical faults (such as motor winding problems, short circuits, overloads, etc.) may not be clearly reflected in the vibration signal. This will cause some fault modes to be unable to be identified, especially when the electrical fault does not cause significant mechanical vibration. On the other hand, vibration signals are easily interfered by noise, especially in industrial environments. There may be other noises in the operating environment of the crane that interfere with the extraction of motor fault features, resulting in a decrease in the diagnostic accuracy of the deep learning model. If there is no other information source to support the diagnosis, the weak noise resistance of the vibration signal alone may lead to misjudgment. Summary of the invention

[0005] In order to solve the technical problem that the current deep learning fault diagnosis technology based on vibration signals has low diagnostic accuracy during the diagnosis process, the present invention provides a fault diagnosis method and system for a crane hook lifting motor.

[0006] The technical solution provided by the embodiment of the present invention is as follows:

[0007] First aspect:

[0008] An embodiment of the present invention provides a method for diagnosing a fault of a crane hook lifting motor, comprising:

[0009] S1: Obtain the vibration signal, current signal and audio signal of the crane hook lifting motor;

[0010] S2: converting the one-dimensional vibration signal, current signal and audio signal into a two-dimensional vibration characteristic graph, current characteristic graph and audio characteristic graph;

[0011] S3: normalizing the vibration characteristic graph, the current characteristic graph, and the audio characteristic graph;

[0012] S4: using the normalized vibration feature map as the R channel, the normalized current feature map as the G channel, and the normalized audio feature map as the B channel to generate a color feature map;

[0013] S5: constructing a convolutional neural network with a cascade structure, wherein the convolutional neural network comprises: an input layer, a first convolutional unit, a second convolutional unit, a third convolutional unit, and a fourth convolutional unit connected in sequence;

[0014] S6: Inputting the color feature map into the input layer of the convolutional neural network;

[0015] S7: extracting a first feature map, a second feature map, a third feature map, and a fourth feature map respectively through the first convolution unit, the second convolution unit, the third convolution unit, and the fourth convolution unit;

[0016] S8: Perform fault diagnosis by the fourth characteristic map alone to obtain a first diagnosis result, perform fault diagnosis by cascading and fusing the third characteristic map and the fourth characteristic map to obtain a second diagnosis result, perform fault diagnosis by cascading and fusing the second characteristic map, the third characteristic map and the fourth characteristic map to obtain a third diagnosis result, and perform fault diagnosis by cascading and fusing the first characteristic map, the second characteristic map, the third characteristic map and the fourth characteristic map to obtain a fourth diagnosis result;

[0017] S9: Determine a final diagnosis result according to the first diagnosis result, the second diagnosis result, the third diagnosis result and the fourth diagnosis result.

[0018] Second aspect:

[0019] An embodiment of the present invention provides a fault diagnosis system for a crane hook lifting motor, comprising:

[0020] processor;

[0021] A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the fault diagnosis method for the crane hook lifting motor as described in the first aspect is implemented.

[0022] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0023] In the present invention, by collecting the vibration signal, current signal and audio signal of the crane hook lifting motor, it is possible to capture the anomalies in the motor operation state from different angles, reduce the interference of environmental noise, and map the vibration signal, current signal and audio signal into two-dimensional feature maps respectively, and fuse them into color feature maps through RGB three channels. This representation form not only retains the characteristics of different signals, but also enables the convolutional neural network to more easily extract useful information from these feature maps, thereby improving the accuracy of the convolutional neural network in diagnosing lifting motor faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0025] Figure 1 A schematic flow chart of a method for diagnosing a fault of a crane hook lifting motor provided by an embodiment of the present invention;

[0026] Figure 2 A structural schematic diagram of a method for diagnosing a fault of a crane hook lifting motor provided by an embodiment of the present invention;

[0027] Figure 3 A schematic structural diagram of a fault diagnosis system for a crane hook lifting motor provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0028] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0029] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0030] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0031] Reference Manual Attached Figure 1, shows a flow chart of a method for diagnosing a fault of a crane hook lifting motor provided by an embodiment of the present invention.

[0032] Reference Manual Attached Figure 2 , showing a structural schematic diagram of a method for diagnosing a fault of a crane hook lifting motor provided by an embodiment of the present invention.

[0033] The embodiment of the present invention provides a method for diagnosing a fault of a crane hook lifting motor, which can be implemented by a fault diagnosis device for the crane hook lifting motor, and the fault diagnosis device for the crane hook lifting motor can be a terminal or a server. The processing flow of the method for diagnosing a fault of a crane hook lifting motor can include the following steps:

[0034] S1: Obtain the vibration signal, current signal and audio signal of the crane hook lifting motor.

[0035] S2: Convert the one-dimensional vibration signal, current signal and audio signal into a two-dimensional vibration characteristic graph, current characteristic graph and audio characteristic graph.

[0036] In a possible implementation, S2 specifically includes sub-steps S201 to S203:

[0037] S201: Convert the one-dimensional vibration signal into a two-dimensional vibration characteristic graph through short-time Fourier transform, wavelet transform or Hilbert-Huang transform.

[0038] Among them, Short-Time Fourier Transform (STFT) is a common method for converting one-dimensional signals into two-dimensional time-frequency diagrams. It divides the signal into multiple short time periods, performs Fourier transform on each time period, and generates a two-dimensional feature diagram of frequency variation over time. STFT can capture the dynamic changes of the frequency components of the signal over time and is suitable for analyzing non-stationary signals, especially in vibration signals, where it can show the frequency changes of fault characteristics over time.

[0039] Among them, Wavelet Transform (WT) is a tool for analyzing the time and frequency characteristics of a signal simultaneously. Unlike Fourier Transform, Wavelet Transform uses a variable time window to decompose the signal, and can adaptively use a smaller time window in the high frequency band and a larger time window in the low frequency band, so as to better capture the instantaneous characteristics of the signal. Wavelet Transform is particularly suitable for processing non-stationary signals and mutation signals, and is often used to extract local anomalies in vibration signals in fault diagnosis.

[0040] Among them, Hilbert-Huang Transform (HHT) is a time-frequency analysis method specifically used to analyze nonlinear and non-stationary signals. It first decomposes the complex signal into several intrinsic mode functions (IMFs) through empirical mode decomposition (EMD), and then performs Hilbert transform on each IMF to generate the instantaneous frequency and energy distribution diagram of the signal. HHT can accurately describe the local characteristics and instantaneous changes of the signal, and is suitable for capturing early signs of faults in vibration signals.

[0041] Those skilled in the art may select one or more methods from short-time Fourier transform, wavelet transform or Hilbert-Huang transform to convert one-dimensional data into two-dimensional data according to actual needs. The present invention does not limit the specific conversion method.

[0042] S202: Convert the one-dimensional current signal into a two-dimensional current characteristic diagram through short-time Fourier transform, wavelet transform or Hilbert-Huang transform.

[0043] S203: Convert the one-dimensional audio signal into a two-dimensional audio feature graph by extracting the Mel-spectrogram.

[0044] Among them, the Mel-spectrogram is a common method for converting audio signals from the time domain to a two-dimensional frequency domain image, which is particularly suitable for audio processing and speech signal analysis. It performs a short-time Fourier transform (STFT) on the audio signal, applies a Mel filter to the spectrum obtained by the STFT, and maps the frequency to the Mel scale. By taking the logarithm of the filtered result and converting it into a logarithmic Mel-spectrogram, it reflects the energy distribution at different frequencies and time points.

[0045] S3: normalizing the vibration characteristic graph, the current characteristic graph, and the audio characteristic graph.

[0046] In a possible implementation, S3 specifically includes: performing linear normalization processing on the vibration characteristic graph, the current characteristic graph, and the audio characteristic graph, and normalizing the values ​​to between 0 and 255:

[0047]

[0048] in, represents the normalized eigenvalue, X represents the eigenvalue before normalization, X max Indicates the maximum value, X min Indicates the minimum value.

[0049] In the present invention, normalizing the numerical value to between 0 and 255 can facilitate the subsequent conversion of color images.

[0050] S4: Generate a color feature map by using the normalized vibration feature map as the R channel, the normalized current feature map as the G channel, and the normalized audio feature map as the B channel.

[0051] In the present invention, by mapping the vibration feature map, current feature map and audio feature map to the RGB channels to generate a color feature map, the fusion and feature enhancement of multiple signal sources can be achieved. This method can not only improve the feature extraction capability and fault diagnosis accuracy of the model, but also capture complex fault modes through multi-channel fusion, thereby improving the generalization and efficiency of the model.

[0052] S5: Construct a convolutional neural network with a cascade structure.

[0053] Among them, the convolutional neural network includes: an input layer, a first convolution unit, a second convolution unit, a third convolution unit and a fourth convolution unit connected in sequence.

[0054] Optionally, four convolution layers are arranged inside the first convolution unit, the second convolution unit, the third convolution unit and the fourth convolution unit.

[0055] The number of channels of the convolutional layer in the first convolutional unit is 64 and the length is 64.

[0056] The number of channels of the convolutional layer in the second convolutional unit is 128 and the length is 32.

[0057] The number of channels of the convolutional layer in the third convolutional unit is 256 and the length is 16.

[0058] The number of channels of the convolutional layer in the fourth convolutional unit is 512 and the length is 8.

[0059] In the present invention, as the layers of convolutional units gradually go deeper, the number of channels (feature dimensions) increases from 64 to 512. This means that the features extracted in the initial layers are relatively basic (such as edges, local changes, etc.), while in the deeper layers, the network can extract more abstract and complex patterns (such as global structures or high-level features). This hierarchical feature abstraction method helps the model capture complex fault modes, especially for multi-dimensional feature analysis in motor fault diagnosis.

[0060] Furthermore, as the layers of convolutional units gradually deepen, the length of each convolutional unit gradually decreases from 64 to 8. The reduction in the size of the feature map means that the network gradually focuses on important feature information. Through convolution and pooling operations, the network can reduce redundant local information, highlight global context and key features, and thus retain the most discriminative information in a more compact representation. This helps to focus on core features and reduce interference information when diagnosing faults.

[0061] Optionally, the convolutional neural network also includes: a dilated convolution unit, and the dilated convolution unit is located between the input layer and the first convolution unit.

[0062] The dilated convolution unit is used to increase the receptive field, extract the dilated feature map from the color feature map, and input the dilated feature map into the first convolution unit:

[0063]

[0064] in, y represents the dilated feature map, y j Represents the position in the expanded feature map j The value at j Represents the position index of the feature map, x represents the input feature map, x ( j + rk ) represents the input feature map j + rk The value at k Represents the position index of the convolution kernel, r represents the expansion rate, w Represents weight.

[0065] Furthermore, in the dilated convolution process, by convolving the input with multi-scale convolution kernels, feature maps at different scales can be obtained. Since directly applying convolution kernels of different sizes may result in computational waste, here we insert a convolution kernel value between two convolution kernel values ​​in each dimension. r -1 zero element is used to upsample the convolution kernel instead of directly scaling the convolution kernel. In this way, the receptive fields of each scale convolution kernel will not overlap with each other, thus effectively avoiding computational waste. On the other hand, since scaling is performed by inserting zero elements into the original convolution kernel, the risk of overfitting will not increase.

[0066] In the present invention, by introducing the dilated convolution unit, the network can increase the receptive field, extract multi-scale features, and retain detail information without increasing the computational complexity, thereby improving the detection capability of complex fault modes. This approach not only enhances the adaptability to non-stationary signals, but also improves the computational efficiency and generalization capability of the model, making fault diagnosis more accurate and stable.

[0067] S6: Input the color feature map to the input layer of the convolutional neural network.

[0068] S7: extracting a first feature map, a second feature map, a third feature map, and a fourth feature map respectively through the first convolution unit, the second convolution unit, the third convolution unit, and the fourth convolution unit.

[0069] Optionally, the first convolution unit includes a first convolution layer, a second convolution layer, a third convolution layer, and a fourth convolution layer connected in sequence, and the first convolution unit extracts the first feature map in a specific manner as follows:

[0070] The input feature map is convolved through the first convolution layer to obtain the first intermediate feature map, which is then input into the second convolution layer.

[0071] The input first intermediate feature map is convolved through the second convolution layer to obtain a second intermediate feature map, which is then input into the third convolution layer.

[0072] The input second intermediate feature map is convolved through the third convolution layer to obtain a third intermediate feature map, which is then input into the fourth convolution layer.

[0073] The third intermediate feature map is convolved through the fourth convolutional layer to obtain a fourth intermediate feature map.

[0074] The first intermediate feature map, the second intermediate feature map, the third intermediate feature map, and the fourth intermediate feature map are fused to obtain a first feature map.

[0075] In the present invention, in the convolutional neural network, the deep convolutional layer can often capture more abstract global features, but this sometimes leads to the loss of low-level detail information. By fusing the intermediate feature maps of each layer, it can be ensured that the low-level detail information (such as local failure modes) will not be completely discarded. The low-level features provide more local details, while the high-level features provide more abstract global information. By fusing these different levels of information together, a more comprehensive and detailed feature map can be generated.

[0076] In a possible implementation, fusing the first intermediate feature map, the second intermediate feature map, the third intermediate feature map, and the fourth intermediate feature map to obtain the first feature map specifically includes:

[0077] Through the first fusion unit, the intermediate feature maps output from different levels are initially fused:

[0078]

[0079] in, Z 1 represents the preliminary fusion feature map of the first level,R 1 represents the first fusion unit, Y 1 represents the first intermediate feature map, none represents a blank image, Z i Indicates i The initial fusion feature map of the level, Z i-1 Indicates i -1 level of preliminary fusion feature map, Y i Indicates i An intermediate feature map, .

[0080] It should be noted that the intermediate feature map of the current layer is fused with the preliminary fused feature map of the upper layer to generate a richer preliminary fused feature map.

[0081] Through the full convolution unit, the preliminary fusion feature maps of each level are convolved to obtain the full convolution feature maps of each level:

[0082]

[0083] in, U i Indicates i The fully convolutional feature map of the layer, FConv represents the fully convolutional unit, .

[0084] In the present invention, the fully convolutional unit enhances the representation capability of the feature map, helps capture features of different scales and levels, and ensures that the feature map contains sufficient information for subsequent high-level fusion and final fault diagnosis.

[0085] Through the second fusion unit, the fully convolutional feature maps at different levels are fused twice:

[0086]

[0087] in, V 1 represents the secondary fusion feature map of the first level, R 2 represents the second fusion unit, U 1 represents the first-level fully convolutional feature map, V i Indicates i The secondary fusion feature map of the level, V i-1 Indicates i -1 level secondary fusion feature map, .

[0088] It should be noted that the preliminary fused feature map of the current layer is fused with the secondary fused feature map of the upper layer to generate a richer secondary fused feature map.

[0089] In the present invention, in the process of feature extraction and fusion, primary fusion and advanced fusion are performed layer by layer to prevent the rapid loss of information. The intermediate feature maps of each layer can participate in the final feature generation process through the fusion unit, avoiding the premature loss of features due to convolution or pooling operations.

[0090] The secondary fusion feature maps of each level are convolved to obtain the first feature map:

[0091]

[0092] in, F 1 represents the first feature map, Conv Represents convolution processing.

[0093] In the present invention, the final convolution processing is performed on the feature map after high-level fusion to further compress and optimize the features and extract the most representative key information.

[0094] Furthermore, how the second convolution unit extracts the second feature map, how the third convolution unit extracts the third feature map, and how the fourth convolution unit extracts the fourth feature map can all refer to the way in which the first convolution unit extracts the first feature map. To avoid repetition, the present invention will not go into details.

[0095] S8: Perform fault diagnosis solely through the fourth characteristic map to obtain a first diagnostic result, perform fault diagnosis after cascading and fusing the third characteristic map and the fourth characteristic map to obtain a second diagnostic result, perform fault diagnosis after cascading and fusing the second characteristic map, the third characteristic map and the fourth characteristic map to obtain a third diagnostic result, and perform fault diagnosis after cascading and fusing the first characteristic map, the second characteristic map, the third characteristic map and the fourth characteristic map to obtain a fourth diagnostic result.

[0096] In a possible implementation, S8 specifically includes sub-steps S801 to S804:

[0097] S801: Perform fault diagnosis alone through the fourth characteristic graph to obtain a first diagnosis result:

[0098]

[0099] in, P 1 represents the first diagnosis result, Softmax represents the Softmax activation function, W 1 represents the first weight matrix, F 4 represents the fourth characteristic graph, b1 represents the first bias term, p 1i Indicates that the first diagnosis result belongs to i The probability value of class failure, n Indicates the total number of fault categories.

[0100] S802: Perform fault diagnosis by cascading and fusing the third feature map and the fourth feature map to obtain a second diagnosis result:

[0101]

[0102] in, P 2 indicates the second diagnosis result, W 2 represents the second weight matrix, F 3 represents the third feature map, represents element-by-element addition, represents the fourth feature map after upsampling. Upsampling is used to make the feature maps have the same size. b 2 represents the second bias term, p 2i Indicates that the second diagnosis result belongs to i The probability value of class failure.

[0103] It should be noted that in multiple fusions, the size of the feature map is adjusted through upsampling operations so that feature maps at different levels can be added element by element. This operation ensures the consistency of the feature maps of each layer in the spatial dimension when the features are fused, avoiding information loss caused by size mismatch.

[0104] S803: Perform fault diagnosis by cascading and fusing the second feature map, the third feature map, and the fourth feature map to obtain a third diagnosis result:

[0105]

[0106] in, P 3 indicates the third diagnosis result, W 3 represents the third weight matrix, F 2 represents the second feature map, represents the third feature map after upsampling, b 3 represents the third bias term, p 3i Indicates that the third diagnosis result belongs to i The probability value of class failure.

[0107] S804: Perform fault diagnosis by cascading and fusing the first feature map, the second feature map, the third feature map, and the fourth feature map to obtain a fourth diagnosis result:

[0108]

[0109] in, P 4 indicates the fourth diagnosis result, W 4 represents the fourth weight matrix, F 1 represents the first feature map, represents the second feature map after upsampling, b 4 represents the fourth bias term, p 4i Indicates that the fourth diagnosis result belongs to i The probability value of class failure.

[0110] In the present invention, in the convolutional neural network, as the number of network layers increases, the resolution of the feature map tends to gradually decrease, which may cause some detail information to be lost. By performing multiple diagnoses at different levels and fusing earlier feature maps, the loss of low-level information caused by convolution and pooling operations can be effectively avoided.

[0111] S9: Determine a final diagnosis result according to the first diagnosis result, the second diagnosis result, the third diagnosis result and the fourth diagnosis result.

[0112] In a possible implementation, S9 specifically includes sub-steps S901 and S902:

[0113] S901: Fusion of the first diagnosis result, the second diagnosis result, the third diagnosis result, and the fourth diagnosis result:

[0114]

[0115] in, p i Indicates that it belongs to i The probability value of class failure.

[0116] S902: The fault category with the largest probability value is taken as the final diagnosis result of this fault detection.

[0117] In the present invention, by fusing the probabilities of the diagnosis results at different levels and selecting the maximum probability as the final diagnosis result, it is possible to effectively integrate the feature information at multiple levels, improve the accuracy, robustness and robustness of the diagnosis, while maintaining a simple and efficient implementation. Through probability fusion, the error caused by insufficient information at a single level can be further reduced, making the diagnosis system more stable and reliable.

[0118] As for the training method of convolutional neural network, the gradient descent method is currently often used to perform overall training on the entire convolutional neural network. However, in a convolutional neural network with a large number of layers, the gradient update speed of the bottom layer is slow, which affects the overall training effect of the network. Therefore, the present invention proposes a new progressive training method.

[0119] First, the gradient descent method can be used to optimize the fourth convolution module separately according to the first detection result. Specifically, a mean square error loss function is constructed, and then the network parameters are updated according to the following formula:

[0120]

[0121] in, i k+1 Indicates k +Model parameters at 1 iteration, i k Indicates k The network parameters at the iteration, or k Indicates k The adaptive learning rate at the iteration, Represents the mean square error loss function J For model parameters i k The gradient of β represents the momentum coefficient.

[0122] It should be noted that the momentum term can help the model avoid falling into the local optimum and smooth the optimization path by considering the previous update direction. The adaptive learning rate method can automatically adjust the learning rate according to the size of the gradient, ensuring that the learning rate is large in the early stage of training to accelerate convergence, and small when close to convergence to refine the search.

[0123] Furthermore, the adaptive learning rate is calculated as:

[0124]

[0125] in, or k Indicates k The adaptive learning rate at the iteration, or 0 represents the initial learning rate, Oil mist concentration prediction loss function J For model parameters i u The gradient of , μ represents a hyperparameter that prevents the denominator from being zero.

[0126] In the present invention, by calculating the cumulative sum of squares of the current gradient, the adaptive learning rate can be dynamically adjusted according to the actual situation during the training process. This means that in the early stage of training, when the gradient changes greatly, the learning rate will be higher, thereby accelerating the convergence speed; in the later stage of training, when the model is close to convergence, the learning rate will be reduced, thereby preventing oscillation and instability caused by excessive update step size. The adaptive learning rate can not only improve the training efficiency and stability of the model, but also improve the prediction accuracy, making the model perform better in complex practical application scenarios.

[0127] After the optimization of the fourth convolution module is completed, the third convolution module and the fourth convolution module are optimized according to the second detection result. The optimization process can refer to the above content of separately optimizing the fourth convolution module according to the first detection result. To avoid repetition, the present invention will not go into details.

[0128] After the optimization of the third convolution module and the fourth convolution module is completed, the second convolution module, the third convolution module and the fourth convolution module are optimized according to the third detection result. The optimization process can refer to the above content of separately optimizing the fourth convolution module according to the first detection result. To avoid repetition, the present invention will not go into details.

[0129] After the optimization of the second convolution module, the third convolution module and the fourth convolution module is completed, the first convolution module, the second convolution module, the third convolution module and the fourth convolution module are optimized according to the fourth detection result. The optimization process can refer to the above content of separately optimizing the fourth convolution module according to the first detection result. To avoid repetition, the present invention will not go into details.

[0130] After the optimization of the first convolution module, the second convolution module, the third convolution module and the fourth convolution module is completed, the remaining part of the convolutional neural network can be optimized according to the final detection result. The optimization process can refer to the above content of optimizing the fourth convolution module according to the first detection result. To avoid repetition, the present invention will not repeat it.

[0131] Finally, the training of the entire convolutional neural network is completed.

[0132] In the present invention, this layer-by-layer modular optimization training method can effectively improve the stability of training, alleviate gradient problems, and enhance feature expression capabilities by gradually optimizing from the last layer of convolutional modules to the front layers, while reducing the risk of model parameter conflicts during training. This method can also improve training efficiency, avoid overfitting, and ultimately ensure that network performance is improved through global optimization. Therefore, this progressive optimization method is very suitable for the training of deep convolutional neural networks, especially in complex tasks such as fault diagnosis.

[0133] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0134] In the present invention, by collecting the vibration signal, current signal and audio signal of the crane hook lifting motor, it is possible to capture the anomalies in the motor operation state from different angles, reduce the interference of environmental noise, and map the vibration signal, current signal and audio signal into two-dimensional feature maps respectively, and fuse them into color feature maps through RGB three channels. This representation form not only retains the characteristics of different signals, but also enables the convolutional neural network to more easily extract useful information from these feature maps, thereby improving the accuracy of the convolutional neural network in diagnosing lifting motor faults.

[0135] Reference Manual Attached Figure 3 , showing a structural schematic diagram of a fault diagnosis system for a crane hook lifting motor provided by the present invention.

[0136] The present invention also provides a fault diagnosis system 20 for a crane hook lifting motor, comprising:

[0137] Processor 201;

[0138] The memory 202 stores computer-readable instructions, and when the computer-readable instructions are executed by the processor 201, the fault diagnosis method for the crane hook lifting motor as described in the method embodiment is implemented.

[0139] The crane hook lifting motor fault diagnosis system 20 provided in the present invention can execute the above-mentioned crane hook lifting motor fault diagnosis method and achieve the same or similar technical effects. To avoid repetition, the present invention will not go into details.

[0140] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0141] In the present invention, by collecting the vibration signal, current signal and audio signal of the crane hook lifting motor, it is possible to capture the anomalies in the motor operation state from different angles, reduce the interference of environmental noise, and map the vibration signal, current signal and audio signal into two-dimensional feature maps respectively, and fuse them into color feature maps through RGB three channels. This representation form not only retains the characteristics of different signals, but also enables the convolutional neural network to more easily extract useful information from these feature maps, thereby improving the accuracy of the convolutional neural network in diagnosing lifting motor faults.

[0142] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

[0143] There are a few points to note:

[0144] (1) The drawings of the embodiments of the present invention only involve structures related to the embodiments of the present invention. Other structures may refer to conventional designs.

[0145] (2) For the sake of clarity, in the drawings used to describe the embodiments of the present invention, the thickness of the film or region is exaggerated or reduced, that is, these drawings are not drawn according to the actual scale. It is understood that when an element such as a film, a region or a substrate is referred to as being "on" or "under" another element, the element may be "directly" "on" or "under" the other element or there may be intermediate elements.

[0146] (3) In the absence of conflict, the embodiments of the present invention and the features therein may be combined with each other to obtain new embodiments.

[0147] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. The protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for diagnosing a fault of a crane hook lifting motor, characterized in that: include: S1: Obtain the vibration signal, current signal and audio signal of the crane hook lifting motor; S2: converting the one-dimensional vibration signal, current signal and audio signal into a two-dimensional vibration characteristic graph, current characteristic graph and audio characteristic graph; S3: normalizing the vibration characteristic graph, the current characteristic graph, and the audio characteristic graph; S4: using the normalized vibration feature map as the R channel, the normalized current feature map as the G channel, and the normalized audio feature map as the B channel to generate a color feature map; S5: constructing a convolutional neural network with a cascade structure, wherein the convolutional neural network comprises: an input layer, a first convolutional unit, a second convolutional unit, a third convolutional unit, and a fourth convolutional unit connected in sequence; S6: Inputting the color feature map into the input layer of the convolutional neural network; S7: extracting a first feature map, a second feature map, a third feature map, and a fourth feature map respectively through the first convolution unit, the second convolution unit, the third convolution unit, and the fourth convolution unit; S8: Perform fault diagnosis by the fourth characteristic map alone to obtain a first diagnosis result, perform fault diagnosis by cascading and fusing the third characteristic map and the fourth characteristic map to obtain a second diagnosis result, perform fault diagnosis by cascading and fusing the second characteristic map, the third characteristic map and the fourth characteristic map to obtain a third diagnosis result, and perform fault diagnosis by cascading and fusing the first characteristic map, the second characteristic map, the third characteristic map and the fourth characteristic map to obtain a fourth diagnosis result; S9: Determine a final diagnosis result according to the first diagnosis result, the second diagnosis result, the third diagnosis result and the fourth diagnosis result.

2. The fault diagnosis method for a crane hook lifting motor according to claim 1, characterized in that: The S2 specifically includes: S201: converting the one-dimensional vibration signal into a two-dimensional vibration characteristic graph through short-time Fourier transform, wavelet transform or Hilbert-Huang transform; S202: converting the one-dimensional current signal into a two-dimensional current characteristic graph by short-time Fourier transform, wavelet transform or Hilbert-Huang transform; S203: Convert the one-dimensional audio signal into a two-dimensional audio feature graph by extracting the Mel-spectrogram.

3. The fault diagnosis method for a crane hook lifting motor according to claim 1, characterized in that: The S3 is specifically: The vibration characteristic graph, the current characteristic graph, and the audio characteristic graph are linearly normalized to normalize the values ​​to between 0 and 255.

4. The fault diagnosis method for a crane hook lifting motor according to claim 1, characterized in that: The convolutional neural network further includes: a dilated convolution unit, wherein the dilated convolution unit is located between the input layer and the first convolution unit; The dilated convolution unit is used to increase the receptive field, extract a dilated feature map from the color feature map, and input the dilated feature map into the first convolution unit.

5. The fault diagnosis method for a crane hook lifting motor according to claim 1, characterized in that: The first convolution unit, the second convolution unit, the third convolution unit and the fourth convolution unit are each provided with four convolution layers; The number of channels of the convolution layer in the first convolution unit is 64 and the length is 64; The number of channels of the convolution layer in the second convolution unit is 128 and the length is 32; The number of channels of the convolution layer in the third convolution unit is 256 and the length is 16; The number of channels of the convolution layer in the fourth convolution unit is 512 and the length is 8.

6. The fault diagnosis method for a crane hook lifting motor according to claim 5, characterized in that: The first convolution unit includes a first convolution layer, a second convolution layer, a third convolution layer, and a fourth convolution layer connected in sequence. The first convolution unit extracts the first feature map in the following specific manner: Performing convolution processing on the input feature map through the first convolution layer to obtain a first intermediate feature map, and inputting it to the second convolution layer; Performing convolution processing on the input first intermediate feature map through the second convolution layer to obtain a second intermediate feature map, and inputting the second intermediate feature map to the third convolution layer; Performing convolution processing on the input second intermediate feature map through the third convolution layer to obtain a third intermediate feature map, and inputting it to the fourth convolution layer; Performing convolution processing on the input third intermediate feature map through the fourth convolution layer to obtain a fourth intermediate feature map; The first intermediate feature map, the second intermediate feature map, the third intermediate feature map, and the fourth intermediate feature map are fused to obtain the first feature map.

7. The method for diagnosing a fault of a crane hook lifting motor according to claim 6, characterized in that: The fusing the first intermediate feature map, the second intermediate feature map, the third intermediate feature map, and the fourth intermediate feature map to obtain the first feature map specifically includes: Through the first fusion unit, the intermediate feature maps output from different levels are initially fused: in, Z 1 represents the preliminary fusion feature map of the first level, R 1 represents the first fusion unit, Y 1 represents the first intermediate feature map, none represents a blank image, Z i Indicates i The initial fusion feature map of the level, Z i-1 Indicates i -1 level of preliminary fusion feature map, Y i Indicates i An intermediate feature map, ; Through the full convolution unit, the preliminary fusion feature maps of each level are convolved to obtain the full convolution feature maps of each level: in, U i Indicates i The fully convolutional feature map of the layer, FConv represents the fully convolutional unit, ; Through the second fusion unit, the fully convolutional feature maps at different levels are fused twice: in, V 1 represents the secondary fusion feature map of the first level, R 2 represents the second fusion unit, U 1 represents the first-level fully convolutional feature map, V i Indicates i The secondary fusion feature map of the level, V i-1 Indicates i -1 level secondary fusion feature map, ; The secondary fusion feature maps of each level are convolved to obtain the first feature map: in, F 1 represents the first feature map, Conv Represents convolution processing.

8. The method for diagnosing a fault of a crane hook lifting motor according to claim 6, characterized in that: The S8 specifically includes: S801: Perform fault diagnosis separately through the fourth characteristic graph to obtain a first diagnosis result: in, P 1 represents the first diagnosis result, Softmax represents the Softmax activation function, W 1 represents the first weight matrix, F 4 represents the fourth characteristic map, b 1 represents the first bias term, p 1i Indicates that the first diagnosis result belongs to i The probability value of class failure, n Indicates the total number of fault categories; S802: Perform fault diagnosis by cascading and fusing the third feature map and the fourth feature map to obtain a second diagnosis result: in, P 2 indicates the second diagnosis result, W 2 represents the second weight matrix, F 3 represents the third feature map, represents element-by-element addition, represents the fourth feature map after upsampling. Upsampling is used to make the feature maps have the same size. b 2 represents the second bias term, p 2i Indicates that the second diagnosis result belongs to i The probability value of class failure; S803: Perform fault diagnosis by cascading and fusing the second feature map, the third feature map, and the fourth feature map to obtain a third diagnosis result: in, P 3 indicates the third diagnosis result, W 3 represents the third weight matrix, F 2 represents the second feature map, represents the third feature map after upsampling, b 3 represents the third bias term, p 3i Indicates that the third diagnosis result belongs to i The probability value of class failure; S804: Perform fault diagnosis by cascading and fusing the first feature map, the second feature map, the third feature map, and the fourth feature map to obtain a fourth diagnosis result: in, P 4 indicates the fourth diagnosis result, W 4 represents the fourth weight matrix, F 1 represents the first feature map, represents the second feature map after upsampling, b 4 represents the fourth bias term, p 4i Indicates that the fourth diagnosis result belongs to i The probability value of class failure.

9. The method for diagnosing a fault of a crane hook lifting motor according to claim 8, characterized in that: The S9 specifically includes: S901: Fusion the first diagnosis result, the second diagnosis result, the third diagnosis result, and the fourth diagnosis result: in, p i Indicates that it belongs to i The probability value of class failure; S902: The fault category with the largest probability value is taken as the final diagnosis result of this fault detection.

10. A fault diagnosis system for a crane hook lifting motor, characterized in that: include: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the fault diagnosis method for the crane hook lifting motor according to any one of claims 1 to 9 is implemented.

Images