Intelligent identification method and device for poor lubrication of equipment for equipment integrity management

Abstract

The invention provides an equipment integrity management-oriented equipment poor lubrication intelligent identification method and device, and relates to the technical field of signal processing, and the method comprises the steps: obtaining a vibration signal of coal mine key rotating equipment, inputting the vibration signal into a preprocessing model, and carrying out the noise reduction and standardization processing, and obtaining a standardized signal; wherein the preprocessing model is used for carrying out noise reduction processing on the vibration signal based on a self-adaptive noise reduction strategy of dynamic wavelet packet decomposition; inputting the standardized signal into a state recognition model to obtain the lubrication state of the coal mine key rotating equipment; the state recognition model comprises a multi-scale convolution module, a space-time attention weighting module, a feature fusion module and a classifier; and the state recognition model is used for performing selective enhancement and fusion on the multi-scale signal features of the standardized signals in the time domain and the frequency domain, and performing classification based on the fused signal features to obtain the lubrication state of the coal mine key rotating equipment. The embodiment of the invention can improve the model stability and the state recognition accuracy.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence and data processing technology, and in particular to a method and device for intelligently identifying poor lubrication of equipment oriented to equipment integrity management. Background Art

[0002] With the increasing mechanization and intelligence of coal mines, key rotating equipment such as shearers, scraper conveyors, and hydraulic supports face multiple failure risks, such as insufficient lubrication, increased wear, and structural loosening, under high-intensity, continuous operation. If these failures are not identified promptly, they will seriously threaten coal mine production safety and equipment lifespan. Traditional equipment fault detection methods, which rely on manual inspections or threshold alarms based on empirical rules, suffer from delayed response, inaccurate fault identification, and poor adaptability to complex operating conditions. These methods fail to meet the real-time, intelligent, and precise operation and maintenance requirements of modern coal mines.

[0003] Conventional vibration signal noise reduction methods often rely on fixed thresholds. This can lead to the loss of key signal features when dealing with varying equipment operating conditions and non-steady-state noise, making it difficult to accurately reflect the equipment's actual status. Existing fault diagnosis methods often focus solely on time-domain or frequency-domain features, ignoring the interaction between the two and often failing to capture the coupling relationships between multiple features. Summary of the Invention

[0004] In light of this, the present invention aims to provide a method and apparatus for intelligently identifying equipment lubrication failure for equipment integrity management. This method utilizes dynamic wavelet packet noise reduction to effectively filter out non-steady-state noise while retaining the high-frequency transient impact and harmonic characteristics unique to lubrication failure. This avoids the mis-filtering of high-frequency fault signals and provides high-quality data for subsequent feature extraction. A multi-scale convolution module encompasses features at different time and frequency scales, and combined with a spatiotemporal attention mechanism to enhance key features, it avoids missed detections due to the flooding of fault features.

[0005] In a first aspect, an embodiment of the present invention provides an intelligent identification method for poor lubrication of equipment for equipment integrity management, comprising: Obtain vibration signals of key rotating equipment in coal mines; Inputting the vibration signal into a preprocessing model, performing noise reduction and normalization processing on the vibration signal to obtain a normalized signal; wherein the preprocessing model is used to perform noise reduction processing on the vibration signal based on an adaptive noise reduction strategy based on dynamic wavelet packet decomposition; The standardized signal is input into a state recognition model to obtain the lubrication status of the key rotating equipment in the coal mine; the state recognition model includes a multi-scale convolution module, a spatiotemporal attention weighted module, a feature fusion module and a classifier; the state recognition model is used to selectively enhance and fuse the multi-scale signal features of the standardized signal in the time domain and frequency domain, and classify based on the fused signal features to obtain the lubrication status of the key rotating equipment in the coal mine.

[0006] In a preferred embodiment of the present invention, the above-mentioned inputting of the standardized signal into the state recognition model to obtain the lubrication state of the key rotating equipment of the coal mine includes: Inputting the standardized signal into a multi-scale convolution module to extract multi-scale signal features; wherein the multi-scale convolution module is constructed based on a multi-branch structure of hybrid dilated convolution and adaptive kernel width; Inputting the multi-scale signal features into the spatiotemporal attention weighting module to obtain the time domain attention weight matrix and the frequency domain attention weight matrix corresponding to the multi-scale signal features; Inputting the multi-scale signal features, the time domain attention weight matrix, and the frequency domain attention weight matrix into a feature fusion module, selectively enhancing and fusing the multi-scale signal features in the time domain and the frequency domain to obtain a fused feature matrix; The fused feature matrix is ​​input into a classifier to obtain the lubrication status of the key rotating equipment in the coal mine; wherein the classifier is constructed based on a pseudo-inverse weight matrix and residual convolution.

[0007] In a preferred embodiment of the present invention, the multi-scale signal features are input into the spatiotemporal attention weighting module to obtain the time domain attention weight matrix and the frequency domain attention weight matrix corresponding to the multi-scale signal features, including: Inputting the multi-scale signal features into the time domain attention weight matrix calculation formula to obtain the time domain attention weight matrix; The multi-scale signal features are input into the frequency domain attention weight matrix calculation formula to obtain the frequency domain attention weight matrix.

[0008] In a preferred embodiment of the present invention, the multi-scale signal features, the time domain attention weight matrix, and the frequency domain attention weight matrix are input into a feature fusion module, and the multi-scale signal features are selectively enhanced and fused in the time domain and frequency domain to obtain a fused feature matrix, including: Determining a time-frequency domain gating weight matrix according to the time-domain attention weight matrix and the frequency-domain attention weight matrix; The multi-scale signal features are weighted by the time domain attention weight matrix and the frequency domain attention weight matrix respectively to obtain the weighted time domain attention weight matrix and the weighted frequency domain attention weight matrix; A fusion feature matrix is ​​determined according to the weighted time-domain attention weight matrix, the weighted frequency-domain attention weight matrix, and the time-frequency-domain gating weight matrix.

[0009] In a preferred embodiment of the present invention, the step of inputting the standardized signal into a multi-scale convolution module to extract multi-scale signal features includes: According to the entropy value of the standardized signal, the kernel width of the corresponding dynamic kernel width convolution is determined; Determining a first signal feature according to a kernel width of the dynamic kernel width convolution; Determining a second signal feature according to a preset baseband convolution kernel width; determining a third signal feature according to a preset dilation rate and a preset dilated convolution kernel width; The first signal feature, the second signal feature, and the third signal feature are concatenated to obtain a multi-scale signal feature.

[0010] In a preferred embodiment of the present invention, the vibration signal is input into a preprocessing model, and noise reduction and standardization processing are performed on the vibration signal to obtain a standardized signal, including: Obtaining a dynamic threshold corresponding to the vibration signal; the dynamic threshold refers to the dynamic threshold corresponding to each of multiple sub-signals obtained after wavelet decomposition of the vibration signal; performing noise reduction on the vibration signal according to the dynamic threshold to obtain a noise-reduced vibration signal; The vibration signal after noise reduction is normalized according to a preset sliding window to obtain a standardized signal.

[0011] In a preferred embodiment of the present invention, the step of reducing noise on the vibration signal according to the dynamic threshold to obtain the noise-reduced vibration signal includes: Performing an inverse discrete wavelet transform based on the dynamic threshold and each of the sub-signals to obtain a time-domain vibration signal; Obtaining an adaptive weight corresponding to the vibration signal; The time domain vibration signal is denoised according to the adaptive weight to obtain a denoised vibration signal.

[0012] In a preferred embodiment of the present invention, before inputting the standardized signal into the state recognition model, the process further includes: Get the initial weights of the initial state recognition model; Acquire multiple sample signals and a classification label corresponding to each of the sample signals; Inputting each of the sample signals and the corresponding classification labels into an initial state recognition model, training the initial state recognition model, and obtaining a loss function result of the initial state recognition model; wherein the loss function is determined based on the wavelet basis, the pseudo-inverse Jacobian stability term, and the time-frequency comparison term in the preprocessing model; According to the loss function result, the initial weight is updated to obtain the updated weight; Counting the number of updates of the updated weights, and returning to the step of inputting the sample signal and the corresponding classification label into an initial state recognition model, training the initial state recognition model, and obtaining a loss function result of the initial state recognition model, until the number of updates is greater than or equal to a preset number of updates; The initial state recognition model is set according to the updated weights to obtain a state recognition model.

[0013] In a preferred embodiment of the present invention, the above-mentioned updating of the initial weights according to the loss function result to obtain the updated weights includes: According to the loss function result, the initial weight is updated to obtain a first weight; For each sample signal, obtaining a sample lubrication state corresponding to the output of the initial state recognition model; For each sample signal, determining the fuzzy membership of the sample signal according to the sample lubrication state; The first weight is adjusted according to the fuzzy membership of each of the sample signals to obtain an updated weight.

[0014] In a second aspect, an embodiment of the present invention further provides an intelligent device for identifying poor lubrication of equipment for equipment integrity management, comprising: Signal acquisition module, used to obtain vibration signals of key rotating equipment in coal mines; a preprocessing module, configured to input the vibration signal into a preprocessing model, perform noise reduction and standardization on the vibration signal, and obtain a standardized signal; wherein the preprocessing model is configured to perform noise reduction on the vibration signal using an adaptive noise reduction strategy based on dynamic wavelet packet decomposition; An identification module is used to input the standardized signal into a state recognition model to obtain the lubrication status of the key rotating equipment in the coal mine; the state recognition model includes a multi-scale convolution module, a spatiotemporal attention weighting module, a feature fusion module and a classifier; the state recognition model is used to selectively enhance and fuse the multi-scale signal features of the standardized signal in the time domain and frequency domain, and classify based on the fused signal features to obtain the lubrication status of the key rotating equipment in the coal mine.

[0015] In a third aspect, an embodiment of the present invention further provides an electronic device comprising a processor and a memory, wherein the memory stores computer-executable instructions that can be executed by the processor, and the processor executes the computer-executable instructions to implement the intelligent identification method for poor lubrication of equipment for equipment integrity management of the first aspect mentioned above.

[0016] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions. When the computer-executable instructions are called and executed by a processor, the computer-executable instructions prompt the processor to implement the intelligent identification method for poor lubrication of equipment for equipment integrity management of the above-mentioned first aspect.

[0017] The embodiments of the present invention bring the following beneficial effects: Embodiments of the present invention provide a method, apparatus, device, and medium for intelligently identifying poor lubrication in equipment for equipment integrity management. Dynamic wavelet packet noise reduction effectively filters out non-steady-state noise while retaining the high-frequency transient impact and harmonic characteristics unique to poor lubrication. This avoids the mis-filtering of high-frequency fault signals and provides high-quality data for subsequent feature extraction. A multi-scale convolution module covers features at different time and frequency scales, and combined with a spatiotemporal attention mechanism to enhance key features, it avoids missed detections due to the submersion of fault features.

[0018] Other features and advantages of the present invention will be set forth in the following description, or some features and advantages may be inferred or unambiguously determined from the description, or may be learned by implementing the above-mentioned technology of the present invention.

[0019] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0021] Figure 1 A flowchart of an intelligent identification method for equipment lubrication failure oriented to equipment integrity management provided by an embodiment of the present invention; Figure 2a A flowchart of another method for intelligently identifying poor lubrication of equipment for equipment integrity management provided by an embodiment of the present invention; Figure 2bA schematic diagram of a fundamental frequency convolution feature extraction result provided by an embodiment of the present invention; Figure 2c A schematic diagram of a fixed dilated convolution feature extraction result provided by an embodiment of the present invention; Figure 2d A schematic diagram of feature extraction results of a multi-scale convolution module provided by an embodiment of the present invention; Figure 2e A schematic diagram of a traditional time attention distribution result provided by an embodiment of the present invention; Figure 2f A schematic diagram of a traditional frequency domain attention distribution result provided by an embodiment of the present invention; Figure 2g A schematic diagram of a frequency domain attention weight matrix result provided by an embodiment of the present invention; Figure 2h A schematic diagram of a temporal attention weight matrix result provided by an embodiment of the present invention; Figure 2i A schematic diagram of performance verification results of a dynamic core width adaptation mechanism provided by an embodiment of the present invention; Figure 3a A flowchart of another method for intelligently identifying poor lubrication of equipment for equipment integrity management provided by an embodiment of the present invention; Figure 3b It is a schematic diagram of vibration signal; Figure 3c Schematic diagram of layer 1 wavelet coefficients; Figure 3d Schematic diagram of layer 2 wavelet coefficients; Figure 3e Schematic diagram of layer 3 wavelet coefficients; Figure 3f Schematic diagram of layer 4 wavelet coefficients; Figure 3g Schematic diagram of layer 5 wavelet coefficients; Figure 3h Schematic diagram of layer 6 wavelet coefficients; Figure 3i Schematic diagram of the vibration signal after noise reduction; Figure 3j It is a schematic diagram of standardized signals; Figure 4a A flowchart of another method for intelligently identifying poor lubrication of equipment for equipment integrity management provided by an embodiment of the present invention; Figure 4b Schematic diagram of the optimization of the state recognition model training process with different initialization methods; Figure 5 A schematic diagram of the structure of an intelligent device for identifying poor lubrication of equipment for equipment integrity management provided by an embodiment of the present invention; Figure 6 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0023] With the increasing mechanization and intelligence of coal mines, key rotating equipment such as shearers, scraper conveyors, and hydraulic supports face multiple failure risks, such as insufficient lubrication, increased wear, and structural loosening, under high-intensity, continuous operation. If these failures are not identified promptly, they will seriously threaten coal mine production safety and equipment lifespan. Traditional equipment fault detection methods, which rely on manual inspections or threshold alarms based on empirical rules, suffer from delayed response, inaccurate fault identification, and poor adaptability to complex operating conditions. These methods fail to meet the real-time, intelligent, and precise operation and maintenance requirements of modern coal mines.

[0024] In the prior art, traditional vibration signal noise reduction methods mostly rely on fixed thresholds. When faced with different equipment operating conditions and non-steady-state noise, it is easy to cause the loss of key signal features and fail to accurately reflect the actual status of the equipment. Existing fault diagnosis methods often only focus on time domain or frequency domain features, ignoring the interaction between the two. The present invention can effectively combine the features of the time domain and frequency domain through a dual time-space attention mechanism, enhance the expression of key fault features, and solve the shortcomings of traditional methods in feature modeling. Traditional convolutional neural networks often find it difficult to capture the coupling relationship between multiple features when processing complex time-varying signals, especially when processing transient impacts and long-term wear features in vibration signals.

[0025] Based on this, embodiments of the present invention provide an intelligent identification method for equipment lubrication failure for equipment integrity management. This method, through adaptive noise reduction using dynamic wavelet packet decomposition, effectively removes strong noise (such as mechanical vibration and electromagnetic interference) in complex coal mine environments, retaining weak fault signals (such as the vibration signature of early bearing wear). Compared to traditional fixed threshold methods, this method avoids false filtering of high-frequency fault signals, providing high-quality data for subsequent feature extraction. Through normalization, it eliminates signal amplitude differences between different devices and operating conditions, making the model applicable to multiple devices of the same type (such as the same model of fans in different mines), reducing retraining costs. A multi-scale convolution module covers features at different time and frequency scales. Combined with a spatiotemporal attention mechanism to enhance key features, it accurately locates key fault locations in the time and frequency domains, enhancing the ability to identify the "time-frequency coupled features" caused by poor lubrication and avoiding missed detections due to the overwhelmed fault signatures (such as the high-frequency signature of minor gear pitting). The classifier outputs specific fault types based on the fused global features (e.g., distinguishing between "poor lubrication" and "bearing wear"), avoiding the ambiguous judgments of traditional methods (e.g., manual auscultation) and improving the targeted nature of maintenance (e.g., determining maintenance plans without disassembly and inspection).

[0026] To facilitate understanding of this embodiment, a method for intelligently identifying poor lubrication of equipment for equipment integrity management disclosed in an embodiment of the present invention is first introduced in detail.

[0027] Example 1 The embodiment of the present invention provides an intelligent identification method for equipment lubrication failure oriented to equipment integrity management. Figure 1 This is a flow chart of an intelligent identification method for poor lubrication of equipment for equipment integrity management provided by an embodiment of the present invention. Figure 1 As shown, the intelligent identification method for poor lubrication of equipment for equipment integrity management may include the following steps: Step S101: Acquire vibration signals of key rotating equipment in a coal mine.

[0028] Vibration signals are generated by key rotating equipment in coal mines, such as bearings in the shearer's cutting section and gearboxes in the reducer of a scraper conveyor. Vibration sensors installed on these equipment collect vibration signals generated during operation.

[0029] In this embodiment of the present invention, a triaxial accelerometer can be used as the vibration sensor, mounted on the equipment's bearing seat or gearbox housing. The sampling frequency is set to 10 kHz, covering the equipment's operating fundamental frequency (0-500 Hz) and its harmonics. The vibration signal is collected in real time by an edge computing gateway, preprocessed with an anti-aliasing filter (cutoff frequency 5 kHz), and stored as a raw time-domain waveform in HDF5 format. This raw time-domain waveform contains the normal vibration characteristics of the equipment, but is also mixed with noise. This signal may be affected by factors such as environmental interference and unstable equipment status.

[0030] Step S102: input the vibration signal into a preprocessing model, perform noise reduction and standardization on the vibration signal, and obtain a standardized signal.

[0031] The preprocessing model is used to perform noise reduction on the vibration signal using an adaptive noise reduction strategy based on dynamic wavelet packet decomposition. Vibration signals exhibit multi-band noise interference and dynamic amplitude fluctuations. Conventional fixed-threshold filtering methods cannot adapt to the differences in noise distribution under different operating conditions, resulting in loss of characteristic information.

[0032] An embodiment of the present invention employs an adaptive noise reduction strategy based on dynamic wavelet packet decomposition, combined with a sliding window normalization method, to eliminate dimensional differences and non-stationary noise interference in vibration signals. Specifically, wavelet packet decomposition can be used to decompose the vibration signal into sub-signals of different frequency bands. Adaptive threshold processing is used to remove noise components from each sub-signal (e.g., by returning noise fluctuations below a threshold to zero), and then the noise-reduced signal is reconstructed to retain valid fault-related vibration characteristics. Exemplarily, the adaptive threshold can dynamically adjust the adaptive threshold by counting local noise energy using a sliding window of a preset width, avoiding the incorrect filtering of high-frequency transient impact characteristics by a fixed threshold and addressing the issue of signal amplitude changes caused by load fluctuations in coal mine equipment. For example, when equipment is suddenly loaded, the noise increases, and the threshold is automatically raised to retain useful signals; when the load is stable, the threshold is lowered for fine-grained noise reduction.

[0033] During the decomposition process, the weights of the sub-signals in each frequency band can be dynamically adjusted using the signal-to-noise ratio (SNR). This enhances high-SNR fault signature frequency bands (such as the harmonic frequency bands of poor lubrication) while suppressing low-SNR noise frequency bands. Compared to a fixed threshold, this method can retain high-frequency transient impact signals such as bearing seizures (which are easily misinterpreted as noise by traditional methods) while filtering out low-frequency mechanical interference from coal mine sites.

[0034] The denoised signal is converted into a standardized signal with a mean of 0 and a variance of 1. This eliminates the impact of signal amplitude differences between different devices and under different working conditions (for example, the vibration amplitude of the same device under light load and heavy load is different). This ensures the consistency of subsequent model inputs, makes the signal smoother in the time dimension, and avoids the impact of local abnormal fluctuations (such as transient voltage interference) on subsequent feature extraction.

[0035] Step S103: input the standardized signal into a state recognition model to obtain the lubrication state of the key rotating equipment in the coal mine.

[0036] The state recognition model includes a multi-scale convolution module, a spatiotemporal attention weighting module, a feature fusion module and a classifier; the state recognition model is used to selectively enhance and fuse the multi-scale signal features of the standardized signal in the time domain and frequency domain, and classify based on the fused signal features to obtain the lubrication status of the key rotating equipment in the coal mine.

[0037] This invention uses a convolutional neural network as a state recognition model, comprising a multi-scale convolution module, a spatiotemporal attention weighting module, an attention feature fusion module, and a multi-layer fully connected neural network classifier module. The state recognition model inputs a standardized signal and, through multi-scale feature extraction, spatiotemporal attention enhancement, feature fusion, and classification, outputs the lubrication status of the device (e.g., normal state, insufficient lubrication, grease contamination, abnormal wear, etc.).

[0038] Specifically, the multi-scale convolution module can consist of multiple convolutional layers with different kernel sizes (e.g., 1×3, 1×5, and 1×7), respectively extracting signal features at different time scales. For example, small kernels capture high-frequency, short-term features (e.g., instantaneous shocks), while large kernels capture low-frequency, long-term features (e.g., periodic vibrations caused by rotor imbalance). The spatiotemporal attention weighting module is divided into time-domain attention weighting and frequency-domain attention weighting. The time-domain attention weighting assigns weights to the features output by the multi-scale convolution module at different time points, highlighting signals at the moment of fault occurrence (e.g., sudden vibration changes caused by bearing seizure). The frequency-domain attention weighting assigns weights to the features output by the multi-scale convolution module at different frequency bands, enhancing frequency components related to the fault (e.g., meshing frequency harmonics caused by gear tooth breakage). The feature fusion module fuses the features extracted by the multi-scale convolution after spatiotemporal attention weighting, integrating global features from both the time domain (temporal variation) and the frequency domain (frequency distribution), thus avoiding the one-sidedness of a single feature. The classifier usually uses a fully connected neural network or a softmax layer to output the classification result of the lubrication status (such as normal state, insufficient lubrication, grease contamination, abnormal wear) based on the fused features, and provides a confidence level (such as 95% probability of insufficient lubrication).

[0039] An embodiment of the present invention provides an intelligent identification method for equipment lubrication problems for equipment integrity management. This method uses adaptive noise reduction based on dynamic wavelet packet decomposition to effectively remove strong noise (such as mechanical vibration and electromagnetic interference) in the complex environment of coal mines, while retaining weak fault signals (such as the vibration characteristics of early bearing wear). Compared with traditional fixed threshold methods, this method avoids false filtering of high-frequency fault signals, providing high-quality data for subsequent feature extraction. Through normalization, it eliminates differences in signal amplitudes across different equipment and operating conditions, making the model applicable to multiple pieces of equipment of the same type (such as the same model of fans in different mines), reducing retraining costs. A multi-scale convolution module covers features at different time and frequency scales. Combined with a spatiotemporal attention mechanism to enhance key features, it accurately locates the key locations of faults in the time and frequency domains, enhancing the ability to identify the "time-frequency coupled features" caused by poor lubrication and avoiding missed detections due to the overwhelmed fault features (such as the high-frequency characteristics of minor gear pitting). The classifier outputs specific fault types based on the fused global features (e.g., distinguishing between "poor lubrication" and "bearing wear"), avoiding the ambiguous judgments of traditional methods (e.g., manual auscultation) and improving the targeted nature of maintenance (e.g., determining maintenance plans without disassembly and inspection).

[0040] Example 2 An embodiment of the present invention also provides another intelligent identification method for equipment poor lubrication for equipment integrity management; this method is implemented on the basis of the method in the above embodiment; this method focuses on describing the specific implementation method of inputting the standardized signal into the state recognition model to obtain the lubrication state of the key rotating equipment in the coal mine.

[0041] Figure 2a Another flow chart of intelligent identification of poor lubrication of equipment for equipment integrity management provided by an embodiment of the present invention is as follows: Figure 2a As shown, the intelligent identification method for poor lubrication of equipment for equipment integrity management may include the following steps: Step S201: Acquire vibration signals of key rotating equipment in a coal mine.

[0042] Step S202: input the vibration signal into a preprocessing model, perform noise reduction and standardization on the vibration signal, and obtain a standardized signal.

[0043] Step S203: input the normalized signal into a multi-scale convolution module to extract multi-scale signal features.

[0044] The multi-scale convolution module is constructed based on a multi-branch structure of hybrid dilated convolution and adaptive kernel width.

[0045] The fixed convolution kernels of conventional convolutional neural networks struggle to capture the coupling between sudden transient features and long-term wear characteristics in vibration signals. This paper employs a multi-branch structure that combines dilated convolution with adaptive kernel width convolution. By combining multi-scale convolutions, it extracts signal features at different time scales, thereby achieving multi-level feature fusion of vibration signals.

[0046] Specifically, the standardized signal is input into the multi-scale convolution module to extract multi-scale signal features, including: determining the kernel width of the corresponding dynamic kernel width convolution according to the entropy value of the standardized signal; determining the first signal feature according to the kernel width of the dynamic kernel width convolution; determining the second signal feature according to the preset kernel width of the baseband convolution; determining the third signal feature according to the preset expansion rate and the preset kernel width of the expansion convolution; and splicing the first signal feature, the second signal feature and the third signal feature to obtain the multi-scale signal feature.

[0047] Among them, the multi-scale convolution module is expressed by the following formula:

[0048] in, For multi-scale signal features, the signal features under different convolution kernel sizes are combined, and multi-scale features are coupled, including fundamental frequency harmonics, long-term wear, and transient impact, to solve the problem that traditional single-scale convolution is insufficient in modeling time-varying fault features.

[0049] is the base frequency convolution, used to calculate the second signal feature, is the vibration signal after noise reduction, that is, the standardized signal, is the kernel width of the baseband convolution. For example, Set to 64.

[0050] is the dilated convolution, used to calculate the third signal feature, d is the dilation rate, is the kernel width of the dilated convolution. For example, d is set to , l is the layer number, It is set to 32. It should be noted that the expansion rate increases exponentially with the number of layers, which expands the receptive field to capture long-term wear characteristics, such as the periodic amplitude modulation caused by gearbox wear.

[0051] is a dynamic kernel width convolution, used to calculate the first signal feature, is the kernel width of the dynamic kernel width convolution. The calculation method is expressed as:

[0052] The convolution kernel width can be dynamically adjusted according to the entropy value of the vibration signal. When the entropy value is high, for example, when the complexity of the transient impact signal is high, the kernel width is increased to capture the features of a wider time window, and vice versa, the kernel width is reduced to focus on details. It represents the calculation of the entropy value of the vibration signal, which is used to measure the complexity of the signal. is the kernel width mapping function, and its calculation method is expressed as:

[0053] in, is the hyperbolic tangent function, This is a rounding operation.

[0054] This means concatenating different convolution outputs along the channel dimension to form multi-channel features and obtain multi-scale signal features. During concatenation, to unify the output channel dimensions of the multi-branch structure, this embodiment of the present invention uses 1×1 convolution for feature channel alignment. This 1×1 convolution maps the number of channels in each branch to a unified dimension, ensuring the effectiveness of subsequent feature fusion.

[0055] Step S204: input the multi-scale signal features into a spatiotemporal attention weighting module to obtain a time-domain attention weight matrix and a frequency-domain attention weight matrix corresponding to the multi-scale signal features.

[0056] The time-domain attention weight matrix is ​​used to selectively enhance multi-scale signal features in the time domain. The frequency-domain attention matrix is ​​used to selectively enhance multi-scale signal features in the frequency domain.

[0057] In the process of processing multi-scale signal features, traditional pseudo-inverse learning lacks attention to the local time-frequency characteristics of the vibration signal. The embodiment of the present invention adopts a spatiotemporal dual attention weight matrix to dynamically weight the features through the spatiotemporal attention weight matrix to achieve selective enhancement of key fault features in the time domain and frequency domain.

[0058] Specifically, the multi-scale signal features are input into the spatiotemporal attention weighting module to obtain the time domain attention weight matrix and the frequency domain attention weight matrix corresponding to the multi-scale signal features, including: inputting the multi-scale signal features into the time domain attention weight matrix calculation formula to obtain the time domain attention weight matrix; inputting the multi-scale signal features into the frequency domain attention weight matrix calculation formula to obtain the frequency domain attention weight matrix.

[0059] The calculation formula of the time domain attention weight matrix can be expressed as:

[0060] is the time-domain attention weight matrix, which represents the correlation of each time step in the time series and strengthens the fault-sensitive time points, such as the starting time of the impact event; is the Softmax function; is the scaling factor; is the temporal attention query matrix, which is calculated as follows:

[0061] is the time domain attention query weight matrix, which is a trainable parameter. It is the feature output by the multi-scale convolution module.

[0062] is the time attention key matrix, and the calculation method is expressed as:

[0063] for The transpose of is the time-domain attention key weight matrix, which is a trainable parameter.

[0064] The calculation formula of the frequency domain attention weight matrix can be expressed as:

[0065] is the frequency domain attention weight matrix, which represents the correlation of frequency domain features. The diagonal matrix is ​​used to selectively enhance fault-related frequency bands, such as the characteristic frequency sidebands caused by poor lubrication. Indicates that the matrix is ​​diagonalized, that is, the diagonal elements are retained and the other elements are set to 0; is the Sigmoid activation function; is the frequency domain attention query weight matrix, which is a trainable parameter; It is the multi-scale signal feature output by the multi-scale convolution module.

[0066] In step S205, the multi-scale signal features, the time domain attention weight matrix, and the frequency domain attention weight matrix are input into a feature fusion module, and the multi-scale signal features are selectively enhanced and fused in the time domain and frequency domain to obtain a fused feature matrix.

[0067] Specifically, in the feature fusion model, multi-scale signal features are first weighted using a time-domain attention weight matrix and a frequency-domain attention weight matrix. These weighted multi-scale signal features are then fused to produce a fused feature matrix. This feature fusion, based on the time-domain and frequency-domain attention weight matrices, accounts for the characteristics of equipment lubrication failures. By combining time-domain and frequency-domain signals to reconstruct fused features, the ability to extract non-stationary shocks and harmonic coupling characteristics caused by lubrication failure is enhanced.

[0068] Furthermore, the multi-scale signal features, the time domain attention weight matrix and the frequency domain attention weight matrix are input into a feature fusion module, and the multi-scale signal features are selectively enhanced and fused in the time domain and frequency domain to obtain a fusion feature matrix, including: determining a time-frequency domain gating weight matrix based on the time domain attention weight matrix and the frequency domain attention weight matrix; weighting the multi-scale signal features by the time domain attention weight matrix and the frequency domain attention weight matrix respectively to obtain a weighted time domain attention weight matrix and a weighted frequency domain attention weight matrix; determining a fusion feature matrix based on the weighted time domain attention weight matrix, the weighted frequency domain attention weight matrix and the time-frequency domain gating weight matrix.

[0069] For example, the fusion feature matrix can be calculated by the following formula:

[0070] In order to fuse the feature matrix, feature resonance is achieved through the coupling of frequency domain filtering enhancement and time domain gating, which solves the problem of insufficient modeling of time-frequency coupling features in traditional methods. Multi-scale signal features output by the multi-scale convolution module; is the time domain attention weight matrix; is the frequency domain attention weight matrix; is the Fourier transform function; is the inverse Fourier transform function; The features are converted to the frequency domain using Fourier transform, and the frequency band selective enhancement is achieved through diagonal matrix. The time domain signal is then reconstructed through inverse Fourier transform to achieve frequency domain filtering enhancement. is the ReLU activation function; ⊙ is the Hadamard product operator; ⊗ is the outer product operator.

[0071] It is a gated weight matrix in the time-frequency domain, combined with time-frequency attention, to dynamically modulate the feature channel weights and enhance the representation of non-stationary impact features; It can be calculated using the following formula:

[0072] is the gate weight matrix, which is a trainable parameter; is the time domain attention weight matrix; is the frequency domain attention weight matrix; is the Sigmoid activation function.

[0073] Step S206: input the fused feature matrix into a classifier to obtain the lubrication status of the key rotating equipment in the coal mine.

[0074] In the embodiment of the present invention, in order to verify the effectiveness of the technical solution in the embodiment of the present invention, the following experimental analysis is performed: Figure 2b A schematic diagram of the fundamental frequency convolution feature extraction results provided by an embodiment of the present invention. Figure 2c A schematic diagram of a fixed dilated convolution feature extraction result provided by an embodiment of the present invention. Figure 2d A schematic diagram of the feature extraction results of a multi-scale convolution module provided by an embodiment of the present invention. Figure 2b 、 Figure 2c and Figure 2d is the feature extraction result of the same vibration signal. Figure 2b 、 Figure 2c and Figure 2d As shown in the figure, the time-frequency feature extraction capabilities of different convolution structures are compared through three-dimensional spectrum heat maps, and the effect of the multi-scale convolution module on capturing the coupling characteristics of transient impact and long-term wear in vibration signals is verified. By comparing the characteristic response distributions of fundamental frequency convolution, fixed expansion convolution and multi-scale convolution of the present invention, it can be seen that the fundamental frequency convolution presents Gaussian distribution characteristics in a specific frequency band, but is insufficient to respond to time domain mutations. The fixed expansion convolution has a response in a wide frequency band area, but the time domain resolution is limited. The technical solution of the embodiment of the present invention forms a high-resolution response peak at the impact feature near 1 second in the time domain through dynamic kernel width adjustment and multi-branch fusion, and at the same time exhibits wide-band enhancement characteristics in the frequency band of 200 to 400 Hz, achieving better feature coverage in both time and frequency dimensions, proving that multi-scale convolution can effectively decouple time-frequency coupling characteristics.

[0075] Figure 2e A schematic diagram of a traditional temporal attention distribution result provided by an embodiment of the present invention. Figure 2f A schematic diagram of a traditional frequency-domain attention distribution result provided by an embodiment of the present invention. Figure 2g A schematic diagram of a frequency-domain attention weight matrix result provided by an embodiment of the present invention. Figure 2h A schematic diagram of a time-domain attention weight matrix result provided by an embodiment of the present invention. Figure 2e 、 Figure 2f 、 Figure 2g and Figure 2hThis is the result of applying an attention mechanism to the same vibration signal. By using a dual-stream heat map to visualize the weight distribution of the spatiotemporal attention mechanism, the attention module's effectiveness in enhancing the key characteristics of poor lubrication faults is verified. Comparing the attention distribution in the time and frequency domains of a conventional method and the technical solution provided by the present invention shows that the conventional method exhibits a single Gaussian distribution in the time dimension, focusing only on the main period of the vibration signal. The frequency-domain attention is concentrated in the fundamental frequency band, with insufficient response to harmonic components. The time-domain attention of the technical solution provided by the present invention forms a bimodal distribution at 0.4 seconds (when the impact occurs) and 1.1 seconds (during the continuous wear phase), accurately locating the time-domain position of the fault characteristics. The frequency-domain attention is simultaneously enhanced at the fundamental frequency and triple frequency, effectively capturing the harmonic resonance phenomenon caused by poor lubrication. This demonstrates that the spatiotemporal attention mechanism can achieve dual-dimensional selective enhancement of fault characteristics.

[0076] Figure 2i A schematic diagram of the performance verification results of a dynamic kernel width adaptation mechanism provided by an embodiment of the present invention. Since the multi-scale convolution module adopts dynamic kernel width convolution during the construction process, such as Figure 2i As shown in the figure, the effectiveness of the dynamic kernel width adaptation mechanism is verified by the dual-axis curve, and the performance of the fixed kernel width and the dynamic kernel width under different signal complexities is compared, revealing the adaptability of the kernel width adjustment strategy to feature extraction. The experimental results show that when the signal complexity is low, the dynamic kernel width is automatically reduced to enhance the time domain resolution. When the complexity is high, the kernel width is increased to capture long-term dependent features. The accuracy curve shows that the dynamic kernel width is better than the fixed kernel width in all kinds of complexity ranges, especially in the medium complexity area, the performance is significantly improved. The entropy-driven kernel width adjustment mechanism effectively realizes the dynamic matching of the convolution receptive field and the signal characteristics, proving the adaptability of the adaptive convolution kernel design to complex working conditions.

[0077] The intelligent identification method for equipment lubrication failure for equipment integrity management, provided by an embodiment of the present invention, employs a dual-time and space-time attention weighting mechanism to dynamically and selectively enhance key fault characteristics in both the time and frequency domains. The time attention matrix strengthens the weights of fault-sensitive time points, while the frequency-domain attention focuses on characteristic frequency bands, improving the detection capability of harmonic resonance phenomena caused by poor lubrication. In the multi-scale convolution module, the combination of multi-scale convolution and adaptive kernel width simultaneously captures both sudden transient impact characteristics and long-term wear characteristics in the signal, addressing the inadequate modeling of time-varying fault characteristics by traditional convolutional neural networks and enabling more comprehensive extraction of the time-frequency characteristics of the vibration signal.

[0078] Example 3 An embodiment of the present invention also provides another intelligent identification method for equipment poor lubrication for equipment integrity management; this method is implemented on the basis of the method in the above embodiment; this method focuses on describing the specific implementation method of inputting the vibration signal into a preprocessing model, performing noise reduction and standardization on the vibration signal, and obtaining a standardized signal.

[0079] Figure 3a A flowchart of another method for intelligently identifying poor lubrication of equipment for equipment integrity management provided by an embodiment of the present invention is shown in FIG. Figure 3a As shown, the intelligent identification method for poor lubrication of equipment for equipment integrity management may include the following steps: Step S301: Acquire vibration signals of key rotating equipment in a coal mine.

[0080] Step S302: Acquire a dynamic threshold corresponding to the vibration signal.

[0081] The dynamic threshold refers to the dynamic threshold corresponding to each of the multiple sub-signals (wavelet coefficients) obtained after wavelet decomposition of the vibration signal. Specifically, for non-steady-state noise in the vibration signal, such as amplitude fluctuations caused by changes in equipment load, the dynamic threshold is determined by counting the local noise energy through a sliding window to avoid the incorrect filtering of high-frequency transient impact characteristics by the fixed threshold. The dynamic threshold can be determined using the following formula:

[0082] in, is the standard deviation of the wavelet coefficient of the kth layer at time t in the window [t-5, t+5], N is the window length; k is a positive integer.

[0083] Step S303 : De-noise the vibration signal according to the dynamic threshold to obtain a de-noised vibration signal.

[0084] Specifically, the vibration signal is denoised according to the dynamic threshold to obtain a noise-reduced vibration signal, including: performing an inverse discrete wavelet transform according to the dynamic threshold and each of the sub-signals to obtain a time-domain vibration signal; obtaining an adaptive weight corresponding to the vibration signal; and denoising the time-domain vibration signal according to the adaptive weight to obtain a noise-reduced vibration signal.

[0085] The vibration signal after noise reduction can be expressed by the following formula:

[0086] is the vibration signal after noise reduction, represents the vibration signal after noise reduction at time t, that is, the smoothed signal obtained by performing wavelet packet decomposition and dynamic hard threshold processing on the vibration signal; It is a vibration signal. The vibration signal at time t includes the vibration characteristics of the device and various noise interferences; is the k-th layer wavelet basis function; represents discrete wavelet transform; represents the inverse discrete wavelet transform, which is used to restore the multiple sub-signals obtained after wavelet decomposition into time domain signals and obtain time domain vibration signals; K is the number of layers of wavelet packet decomposition, that is, the number of wavelet basis functions selected in the signal decomposition process; let is v, is a dynamic hard threshold function, the formula of the dynamic hard threshold function is as follows:

[0087] is the dynamic threshold of the kth layer at the tth moment.

[0088] is the adaptive weight at time t, and its calculation method is expressed as:

[0089] Adaptive weights are used to selectively enhance high signal-to-noise ratio sub-bands in the frequency domain, such as the harmonic components of the equipment fundamental frequency, suppress low signal-to-noise ratio sub-bands such as environmental noise, and highlight the harmonic distortion characteristics caused by poor lubrication.

[0090] in, is the smoothing factor; For the Moment Layer signal-to-noise ratio, calculated as:

[0091] For the Moment The energy of the noise signal of the layer; is the first The layer signal is The energy of the moment.

[0092] Step S304 : normalizing the noise-reduced vibration signal according to a preset sliding window to obtain a standardized signal.

[0093] Specifically, a sliding window of fixed length is first set. Each time the signal sequence moves forward, the data points within the window are extracted, centered on the current data point. The mean and standard deviation of all data points within the window are then calculated. The normalized value is then obtained by subtracting the window mean from the current data point and dividing it by the window standard deviation. This process is repeated, with the window sliding sequentially until the end of the signal. This achieves local normalization of each data point within its neighborhood, making the signal more stable in time and facilitating subsequent feature extraction. The preset sliding window length can be the same as the sliding window length set when determining the dynamic threshold.

[0094] Step S305: input the standardized signal into a state recognition model to obtain the lubrication state of the key rotating equipment in the coal mine.

[0095] For example, Figure 3b Schematic diagram of vibration signal. Figure 3c Schematic diagram of layer 1 wavelet coefficients. Figure 3d Schematic diagram of layer 2 wavelet coefficients. Figure 3e Schematic diagram of layer 3 wavelet coefficients. Figure 3f Schematic diagram of layer 4 wavelet coefficients. Figure 3g Schematic diagram of layer 5 wavelet coefficients. Figure 3h is a schematic diagram of the layer 6 wavelet coefficients. Figures 3c to 3h , 6 figures show the following Figure 3b The vibration signal (original signal) shown here shows each layer of coefficients after wavelet packet decomposition. The wavelet transform decomposes the signal into multiple frequency bands. Each layer of wavelet coefficients represents the signal's information within a different frequency range. The low-frequency portion contains the signal's main trends and slow changes, while the high-frequency portion contains detailed information and noise. Wavelet packet decomposition can separate the signal's different frequency band components, helping to identify noise and useful features within the signal. In the noise reduction process, it is usually necessary to remove high-frequency noise components while retaining the low-frequency, primary signal components.

[0096] Figure 3i is a schematic diagram of the vibration signal after noise reduction. Figure 3i As shown in the figure, the noise reduction signal is obtained after wavelet packet decomposition and hard threshold processing. After hard threshold processing, the noise component is suppressed and the main features of the signal are retained. It can be seen that the noise is significantly reduced. By removing the noise component in the signal, the signal becomes smoother, which is convenient for subsequent feature extraction and analysis. The denoised signal retains the normal vibration characteristics of the equipment and removes environmental noise and other interference, which helps to improve the diagnostic accuracy.

[0097] Figure 3j is a schematic diagram of standardized signals. Figure 3jAs shown in Figure 1, the noise reduction signal after sliding window normalization processing, during the normalization process, each data point of the signal is normalized according to its adjacent window data, eliminating the dimensional difference and non-steady-state fluctuation of the signal. The purpose of normalization is to eliminate the influence of dimensional difference or time scale difference in the signal, so that the signals of different equipment or different working conditions can be compared and analyzed under the same standard. In addition, the standardized signal is smoother, which helps to eliminate local abnormal fluctuations, making subsequent feature extraction and model training more stable and efficient.

[0098] The intelligent identification method for equipment lubrication failure for equipment integrity management provided by an embodiment of the present invention uses dynamic wavelet packet decomposition combined with hard threshold processing to perform adaptive noise reduction on vibration signals, dynamically adjusts the threshold to adapt to the signal noise characteristics under different working conditions, can effectively remove non-steady-state noise, retain the vibration characteristics of the equipment, solves the shortcomings of traditional fixed threshold methods under different working conditions, and avoids the false filtering of high-frequency transient impact characteristics.

[0099] Example 4 An embodiment of the present invention also provides another intelligent identification method for equipment poor lubrication for equipment integrity management; this method is implemented on the basis of the method in the above embodiment; this method focuses on describing the specific implementation method of inputting the vibration signal into a preprocessing model, performing noise reduction and standardization on the vibration signal, and obtaining a standardized signal.

[0100] Figure 4a A flowchart of another method for intelligently identifying poor lubrication of equipment for equipment integrity management provided by an embodiment of the present invention is shown in FIG. Figure 4a As shown, the intelligent identification method for poor lubrication of equipment for equipment integrity management may include the following steps: Step S401: Acquire vibration signals of key rotating equipment in a coal mine.

[0101] Step S402: input the vibration signal into a preprocessing model, perform noise reduction and standardization on the vibration signal, and obtain a standardized signal.

[0102] Step S403: Obtain the initial weight of the initial state recognition model.

[0103] The initial state recognition model refers to the state recognition model before training. The initial weight refers to the initial value set for the initial state recognition model. In the embodiment of the present invention, an initialization method based on pseudo-inverse learning is adopted to ensure that the initial weight conforms to the time-frequency characteristics of the data. Specifically, the initial weight can be expressed as follows:

[0104] is the initial weight; Φ is the feature matrix, and , is the first vibration signal after noise reduction, is the Nth vibration signal after noise reduction; is the transpose of Φ; N is the number of sample signals; is the one-hot encoding matrix of the lubrication state; I is the identity matrix; λ is the ridge regression coefficient, such as, For sample signals, domain experts combine time-frequency analysis to determine and label the lubrication status. Examples of lubrication status include normal, insufficient lubrication, grease contamination, and abnormal wear. Labeling uses one-hot encoding to represent multi-class labels, such as [1,0,0,0] for normal and [0,1,0,0] for insufficient lubrication.

[0105] Step S404: Acquire multiple sample signals and a classification label corresponding to each of the sample signals.

[0106] Classification labels refer to the annotation results obtained by annotating vibration signals, also known as one-hot encoding. Domain experts combine time-frequency analysis to determine the lubrication status and label it. Sample signals can be multiple vibration signals collected from key rotating equipment in coal mines over a historical period.

[0107] Step S405 : inputting each of the sample signals and the corresponding classification labels into an initial state recognition model, training the initial state recognition model, and obtaining a loss function result of the initial state recognition model.

[0108] The loss function is determined according to the wavelet basis, the pseudo-inverse Jacobian stability term and the time-frequency comparison term in the preprocessing model.

[0109] Since coal mine equipment failures have multi-scale time-frequency feature coupling and pseudo-inverse weight stability requirements, the embodiment of the present invention adopts multi-domain contrast regularization loss, and through time-frequency contrast learning and pseudo-inverse Jacobian constraints, it realizes feature decoupling and model generalization enhancement, thereby constructing a loss function. Specifically, the loss function can be expressed by the following formula:

[0110] Where, As a feature alignment term in the wavelet domain, the wavelet basis of the denoising process is embedded into the loss function as prior knowledge to achieve end-to-end optimization and feature space consistency constraints, and, The L1 norm is used in the dynamic wavelet packet feature extraction operator to enhance the robustness to abnormal impact features.

[0111] As a pseudo-inverse Jacobian stability term, the numerical instability of the pseudo-inverse weight is addressed by regularizing the Jacobian matrix to improve the iterative convergence, and, Characterizing the Jacobian matrix of the pseudo-inverse weights to the input features, constraining its Frobenius norm can suppress gradient explosion.

[0112] As a time-frequency comparison term, a multimodal contrastive learning strategy is adopted to force the model to establish a time-frequency consistency representation, solving the problem that traditional cross-entropy loss is insufficient in modeling time-frequency coupling features.

[0113] in, is the influence factor of the pseudo-inverse Jacobian stability term, for example, ; is the impact factor of the time-frequency comparison term, for example, ; is the similarity threshold, exemplarily, ; is the loss function of the convolutional neural network; It is a dynamic wavelet packet feature extraction operator, reused in S1 The weights force the network to learn representations that are aligned with the wavelet denoising feature space; For logarithmic functions, the default base is a natural constant; is the L1 norm; is the feature representation of the i-th sample after passing through the convolutional neural network; is the true fault category label of the i-th sample; is the total number of layers of the convolutional neural network; is the symbol of partial derivative; is the Frobenius norm; is the feature representation of the i-th sample in the batch; is the feature representation of the jth sample in the batch; is the similarity scaling factor, exemplarily, ; is the cosine similarity function; is the batch size, that is, the number of sample signals input into the initial state recognition model.

[0114] Specifically, a plurality of sample signals and a classification label corresponding to each sample signal are input into an initial state recognition model, the initial state recognition model is trained, and a loss function result of the initial state recognition model is calculated.

[0115] Step S406: Update the initial weights according to the loss function result to obtain updated weights.

[0116] Since the traditional gradient descent method does not consider the differential impact of the importance of spatiotemporal features on parameter updates, the embodiment of the present invention proposes a parameter update strategy that integrates spatiotemporal attention images, adjusts the gradient update direction through the spatiotemporal attention matrix, and realizes adaptive optimization of the initial weights. Specifically, the initial weights can be updated according to the following formula:

[0117] Where, For the The updated weight obtained by the iteration; For the The updated weight obtained by the iteration; The loss function of the initial state recognition model is about The gradient of the initial weights at iterations; The learning rate of the initial state recognition model, for example, ; is the label matrix.

[0118] Step S407, counting the number of updates of the updated weights, and returning to the step of inputting the sample signal and the corresponding classification label into the initial state recognition model, training the initial state recognition model, and obtaining the loss function result of the initial state recognition model, until the number of updates is greater than or equal to the preset number of updates.

[0119] Specifically, step S405 and step S406 are repeatedly performed until a preset stop iteration condition is met, indicating that the model training is completed. Exemplarily, the preset stop iteration condition is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0120] Step S408: Setting the initial state recognition model according to the updated weights to obtain a state recognition model.

[0121] Specifically, the updated weights are set as parameters of the initial state recognition model to obtain the state recognition model.

[0122] Step S409: input the standardized signal into a state recognition model to obtain the lubrication state of the key rotating equipment in the coal mine.

[0123] Furthermore, according to the loss function result, the initial weight is updated to obtain the updated weight, including: according to the loss function result, the initial weight is updated to obtain the first weight; for each sample signal, the sample lubrication state corresponding to the output of the initial state recognition model is obtained; for each sample signal, the fuzzy membership of the sample signal is determined according to the sample lubrication state; according to the fuzzy membership of each sample signal, the first weight is adjusted to obtain the updated weight.

[0124] The first weight refers to the updated weight obtained after the first iteration in the training process of the initial state recognition model. The fuzzy membership is used to describe the importance of the sample signal to the training of the initial state recognition model.

[0125] To address the problem of equipment operating condition drift in coal mines and mitigate the impact of equipment operating condition drift on model performance, such as the impact of vibration mode shift caused by ambient temperature changes on model performance, during the training of the initial state recognition model, a fuzzy membership can be set for each sample signal after the first iteration. The fuzzy membership can be used to dynamically adjust the weight of the sample signal, thereby increasing the adaptability of the trained state recognition model to different operating conditions and improving the state recognition model's ability to adapt to changing data. It can be understood that for each sample signal, the corresponding fuzzy membership function can be used to determine the corresponding fuzzy membership.

[0126] The fuzzy membership function can be expressed by the following formula:

[0127] Where, For the The fuzzy membership of each sample is dynamically adjusted according to the confidence level of the sample fault state and lubrication state output by the initial state recognition model. The weight of the sample signal with low confidence level (possibly a new working condition) is reduced, that is, the fuzzy membership level is reduced, thereby reducing the error update of the initial state recognition model. is the prediction confidence of the i-th sample; is an exponential function with a natural constant as its base; is the slope factor, exemplarily, ; is the confidence threshold, for example, ; Furthermore, the first weight is incrementally updated according to the fuzzy membership of each sample signal to adjust the first weight and obtain the updated weight, thereby achieving adaptive adjustment of the model, which is expressed as:

[0128]

[0129] Where, is the updated convolutional neural network weight; is the fuzzy membership matrix; is the fuzzy membership of the first sample; is the fuzzy membership of the Nth sample.

[0130] Figure 4b The following is a schematic diagram of the optimization of the state recognition model training process using different initialization methods. Figure 4b As shown in the figure, the optimization effect of pseudo-inverse initialization on the model training process is analyzed by the parameter space contour map, and the parameter update paths of random initialization and pseudo-inverse initialization are compared to reveal the convergence characteristics of different initialization methods in non-convex optimization. The traditional random initialization path shows violent oscillations and repeatedly wanders in the saddle point area of ​​the loss surface. The technical solution of the embodiment of the present invention, the initialization path directly points to the global optimal neighborhood, the trajectory is smooth and converges rapidly, and the parameter space visualization shows that the initial value of the pseudo-inverse weight is located in the low-gradient flat area of ​​the loss surface, which is highly consistent with the time-frequency feature distribution of the vibration signal. Theoretically, it is guaranteed that the optimization process avoids local minimum traps, and the effect of pseudo-inverse initialization on improving the stability of model training is verified.

[0131] The technical solution of the embodiment of the present invention uses pseudo-inverse initialization and pseudo-inverse Jacobian stability terms to optimize the training process of the state recognition model. Pseudo-inverse initialization directly makes the network weights more consistent with the time-frequency characteristics of the vibration signal, avoiding the slow convergence and instability problems that may be caused by traditional random initialization. The pseudo-inverse Jacobian stability term further ensures the stability of the gradient and the convergence of the model during the non-convex optimization process. To address the operating condition drift problem of coal mining equipment, a fuzzy membership function and incremental learning strategy are adopted to dynamically adjust the weights of training samples, enhance the model's adaptability to data under different operating conditions, and reduce the weight of low-confidence samples to reduce model erroneous updates, thereby improving the model's stability and reliability.

[0132] Example 5 Corresponding to the above method embodiment, the embodiment of the present invention provides an intelligent device for identifying poor lubrication of equipment for equipment integrity management. Figure 5 A schematic diagram of the structure of an intelligent device for identifying poor lubrication of equipment for equipment integrity management provided by an embodiment of the present invention is shown as follows: Figure 5 As shown, the intelligent identification device for poor lubrication of equipment for equipment integrity management may include: Signal acquisition module 501, used to acquire vibration signals of key rotating equipment in coal mines; A preprocessing module 502 is configured to input the vibration signal into a preprocessing model, perform noise reduction and normalization on the vibration signal, and obtain a normalized signal; wherein the preprocessing model is configured to perform noise reduction on the vibration signal using an adaptive noise reduction strategy based on dynamic wavelet packet decomposition; The identification module 503 is used to input the standardized signal into the state recognition model to obtain the lubrication status of the key rotating equipment in the coal mine; the state recognition model includes a multi-scale convolution module, a spatiotemporal attention weighting module, a feature fusion module and a classifier; the state recognition model is used to selectively enhance and fuse the multi-scale signal features of the standardized signal in the time domain and frequency domain, and classify based on the fused signal features to obtain the lubrication status of the key rotating equipment in the coal mine.

[0133] An embodiment of the present invention provides an intelligent device for identifying equipment lubrication problems for equipment integrity management. This device uses adaptive noise reduction based on dynamic wavelet packet decomposition to effectively remove strong noise (such as mechanical vibration and electromagnetic interference) in the complex environment of coal mines, while retaining weak fault signals (such as the vibration characteristics of early bearing wear). Compared to traditional fixed threshold methods, this method avoids false filtering of high-frequency fault signals, providing high-quality data for subsequent feature extraction. Through normalization, it eliminates differences in signal amplitudes across different equipment and operating conditions, making the model applicable to multiple pieces of equipment of the same type (such as the same model of fans in different mines), reducing retraining costs. A multi-scale convolutional module covers features at different time and frequency scales. Combined with a spatiotemporal attention mechanism to enhance key features, it accurately locates key fault locations in the time and frequency domains, enhancing the ability to identify the "time-frequency coupled features" caused by poor lubrication and avoiding missed faults due to the overwhelmed fault features (such as the high-frequency characteristics of minor gear pitting). The classifier outputs specific fault types based on the fused global features (e.g., distinguishing between "poor lubrication" and "bearing wear"), avoiding the ambiguous judgments of traditional methods (e.g., manual auscultation) and improving the targeted nature of maintenance (e.g., determining maintenance plans without disassembly and inspection).

[0134] In some embodiments, the identification module 503 is further configured to: Inputting the standardized signal into a multi-scale convolution module to extract multi-scale signal features; wherein the multi-scale convolution module is constructed based on a multi-branch structure of hybrid dilated convolution and adaptive kernel width; Inputting the multi-scale signal features into the spatiotemporal attention weighting module to obtain the time domain attention weight matrix and the frequency domain attention weight matrix corresponding to the multi-scale signal features; Inputting the multi-scale signal features, the time domain attention weight matrix, and the frequency domain attention weight matrix into a feature fusion module, selectively enhancing and fusing the multi-scale signal features in the time domain and the frequency domain to obtain a fused feature matrix; The fused feature matrix is ​​input into a classifier to obtain the lubrication status of the key rotating equipment in the coal mine; wherein the classifier is constructed based on a pseudo-inverse weight matrix and residual convolution.

[0135] In some embodiments, inputting the multi-scale signal features into a spatiotemporal attention weighting module to obtain a time-domain attention weight matrix and a frequency-domain attention weight matrix corresponding to the multi-scale signal features includes: Inputting the multi-scale signal features into the time domain attention weight matrix calculation formula to obtain the time domain attention weight matrix; The multi-scale signal features are input into the frequency domain attention weight matrix calculation formula to obtain the frequency domain attention weight matrix.

[0136] In some embodiments, the multi-scale signal features, the time domain attention weight matrix, and the frequency domain attention weight matrix are input into a feature fusion module, and the multi-scale signal features are selectively enhanced and fused in the time domain and frequency domain to obtain a fused feature matrix, including: Determining a time-frequency domain gating weight matrix according to the time-domain attention weight matrix and the frequency-domain attention weight matrix; The multi-scale signal features are weighted by the time domain attention weight matrix and the frequency domain attention weight matrix respectively to obtain the weighted time domain attention weight matrix and the weighted frequency domain attention weight matrix; A fusion feature matrix is ​​determined according to the weighted time-domain attention weight matrix, the weighted frequency-domain attention weight matrix, and the time-frequency-domain gating weight matrix.

[0137] In some embodiments, the normalized signal is input into a multi-scale convolution module to extract multi-scale signal features, including: According to the entropy value of the standardized signal, the kernel width of the corresponding dynamic kernel width convolution is determined; Determining a first signal feature according to a kernel width of the dynamic kernel width convolution; Determining a second signal feature according to a preset baseband convolution kernel width; determining a third signal feature according to a preset dilation rate and a preset dilated convolution kernel width; The first signal feature, the second signal feature, and the third signal feature are concatenated to obtain a multi-scale signal feature.

[0138] In some embodiments, the pre-processing module 502 is further configured to: Obtaining a dynamic threshold corresponding to the vibration signal; the dynamic threshold refers to the dynamic threshold corresponding to each of multiple sub-signals obtained after wavelet decomposition of the vibration signal; performing noise reduction on the vibration signal according to the dynamic threshold to obtain a noise-reduced vibration signal; The vibration signal after noise reduction is normalized according to a preset sliding window to obtain a standardized signal.

[0139] In some embodiments, performing noise reduction on the vibration signal according to the dynamic threshold to obtain a noise-reduced vibration signal includes: Performing an inverse discrete wavelet transform based on the dynamic threshold and each of the sub-signals to obtain a time-domain vibration signal; Obtaining an adaptive weight corresponding to the vibration signal; The time domain vibration signal is denoised according to the adaptive weight to obtain a denoised vibration signal.

[0140] In some embodiments, the apparatus further comprises: Initialization module, used to obtain the initial weights of the initial state recognition model; A sample acquisition module, configured to acquire a plurality of sample signals and a classification label corresponding to each of the sample signals; a training module, configured to input each of the sample signals and the corresponding classification labels into an initial state recognition model, train the initial state recognition model, and obtain a loss function result of the initial state recognition model; wherein the loss function is determined based on the wavelet basis, the pseudo-inverse Jacobian stability term, and the time-frequency comparison term in the preprocessing model; An updating module, configured to update the initial weights according to the loss function result to obtain updated weights; a feedback module, configured to count the number of updates of the updated weights and return to execute the step of inputting the sample signal and the corresponding classification label into the initial state recognition model, training the initial state recognition model, and obtaining a loss function result of the initial state recognition model, until the number of updates is greater than or equal to a preset number of updates; A setting module is used to set the initial state recognition model according to the updated weight to obtain a state recognition model.

[0141] The device provided in the embodiment of the present invention has the same implementation principle and technical effects as those in the aforementioned method embodiment. For the sake of brief description, for matters not mentioned in the device embodiment, reference can be made to the corresponding content in the aforementioned method embodiment.

[0142] Example 5 The embodiment of the present invention further provides an electronic device for executing the above-mentioned device integrity management-oriented intelligent identification method for poor lubrication of equipment; Figure 6 A structural schematic diagram of an electronic device is shown, which includes a memory 600 and a processor 601, wherein the memory 600 is used to store one or more computer instructions, and the one or more computer instructions are executed by the processor 601 to implement the above-mentioned intelligent identification method for poor lubrication of equipment for equipment integrity management.

[0143] Further, Figure 6 The electronic device shown further includes a bus 602 and a communication interface 603 , and the processor 601 , the communication interface 603 and the memory 600 are connected via the bus 602 .

[0144] The memory 600 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage. The communication connection between the system network element and at least one other network element is achieved through at least one communication interface 603 (which may be wired or wireless), and may use the Internet, a wide area network, a local area network, a metropolitan area network, etc. The bus 602 may be an ISA bus, a PCI bus, or an EISA bus. Buses can be divided into address buses, data buses, control buses, etc. For ease of representation, Figure 6 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0145] The processor 601 may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by hardware integrated logic circuits in the processor 601 or by software instructions. The above processor 601 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present invention. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in conjunction with the embodiments of the present invention can be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium well-known in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The storage medium is located in memory 600, and processor 601 reads the information in memory 600 and, in conjunction with its hardware, completes the steps of the method of the aforementioned embodiment.

[0146] An embodiment of the present invention also provides a computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are called and executed by a processor, the computer-executable instructions prompt the processor to implement the above-mentioned intelligent identification method for poor lubrication of equipment for equipment integrity management. The specific implementation can be found in the method embodiment and will not be repeated here.

[0147] The computer program product for the intelligent identification method of equipment lubrication problems for equipment integrity management provided in an embodiment of the present invention includes a computer-readable storage medium storing non-volatile program code executable by a processor. The instructions included in the program code can be used to execute the method described in the previous method embodiment. For specific implementation, please refer to the method embodiment and will not be repeated here.

[0148] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0149] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. There may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed may be through some communication interface, indirect coupling or communication connection of devices or units, which may be electrical, mechanical or other forms.

[0150] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0151] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0152] If the functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage media include various media capable of storing program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0153] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the above-described embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above-described embodiments within the technical scope disclosed by the present invention, or replace some of the technical features therein with equivalents. Such modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. An intelligent identification method for equipment lubrication failure for equipment integrity management, characterized by: include: Obtain vibration signals of key rotating equipment in coal mines; Inputting the vibration signal into a preprocessing model, performing noise reduction and normalization processing on the vibration signal to obtain a normalized signal; wherein the preprocessing model is used to perform noise reduction processing on the vibration signal based on an adaptive noise reduction strategy based on dynamic wavelet packet decomposition; The standardized signal is input into a state recognition model to obtain the lubrication status of the key rotating equipment in the coal mine; the state recognition model includes a multi-scale convolution module, a spatiotemporal attention weighted module, a feature fusion module and a classifier; the state recognition model is used to selectively enhance and fuse the multi-scale signal features of the standardized signal in the time domain and frequency domain, and classify based on the fused signal features to obtain the lubrication status of the key rotating equipment in the coal mine.

2. The method according to claim 1, characterized in that The step of inputting the standardized signal into a state recognition model to obtain the lubrication state of the key rotating equipment in the coal mine includes: Inputting the standardized signal into a multi-scale convolution module to extract multi-scale signal features; wherein the multi-scale convolution module is constructed based on a multi-branch structure of hybrid dilated convolution and adaptive kernel width; Inputting the multi-scale signal features into the spatiotemporal attention weighting module to obtain the time domain attention weight matrix and the frequency domain attention weight matrix corresponding to the multi-scale signal features; Inputting the multi-scale signal features, the time domain attention weight matrix, and the frequency domain attention weight matrix into a feature fusion module, selectively enhancing and fusing the multi-scale signal features in the time domain and the frequency domain to obtain a fused feature matrix; The fused feature matrix is ​​input into a classifier to obtain the lubrication status of the key rotating equipment in the coal mine; wherein the classifier is constructed based on a pseudo-inverse weight matrix and residual convolution.

3. The method according to claim 2, characterized in that The multi-scale signal features are input into the spatiotemporal attention weighting module to obtain the time domain attention weight matrix and the frequency domain attention weight matrix corresponding to the multi-scale signal features, including: Inputting the multi-scale signal features into the time domain attention weight matrix calculation formula to obtain the time domain attention weight matrix; The multi-scale signal features are input into the frequency domain attention weight matrix calculation formula to obtain the frequency domain attention weight matrix.

4. The method according to claim 2, characterized in that The multi-scale signal features, the time domain attention weight matrix, and the frequency domain attention weight matrix are input into a feature fusion module, and the multi-scale signal features are selectively enhanced and fused in the time domain and frequency domain to obtain a fused feature matrix, including: Determining a time-frequency domain gating weight matrix according to the time-domain attention weight matrix and the frequency-domain attention weight matrix; The multi-scale signal features are weighted by the time domain attention weight matrix and the frequency domain attention weight matrix respectively to obtain the weighted time domain attention weight matrix and the weighted frequency domain attention weight matrix; A fusion feature matrix is ​​determined according to the weighted time-domain attention weight matrix, the weighted frequency-domain attention weight matrix, and the time-frequency-domain gating weight matrix.

5. The method according to claim 2, characterized in that The normalized signal is input into a multi-scale convolution module to extract multi-scale signal features, including: According to the entropy value of the standardized signal, the kernel width of the corresponding dynamic kernel width convolution is determined; Determining a first signal feature according to a kernel width of the dynamic kernel width convolution; Determining a second signal feature according to a preset baseband convolution kernel width; determining a third signal feature according to a preset dilation rate and a preset dilated convolution kernel width; The first signal feature, the second signal feature, and the third signal feature are concatenated to obtain a multi-scale signal feature.

6. The method according to claim 1, wherein The step of inputting the vibration signal into a preprocessing model, performing noise reduction and standardization processing on the vibration signal to obtain a standardized signal comprises: Obtaining a dynamic threshold corresponding to the vibration signal; the dynamic threshold refers to the dynamic threshold corresponding to each of multiple sub-signals obtained after wavelet decomposition of the vibration signal; performing noise reduction on the vibration signal according to the dynamic threshold to obtain a noise-reduced vibration signal; The vibration signal after noise reduction is normalized according to a preset sliding window to obtain a standardized signal.

7. The method according to claim 6, characterized in that The step of reducing noise on the vibration signal according to the dynamic threshold to obtain a noise-reduced vibration signal includes: Performing an inverse discrete wavelet transform based on the dynamic threshold and each of the sub-signals to obtain a time-domain vibration signal; Obtaining an adaptive weight corresponding to the vibration signal; The time domain vibration signal is denoised according to the adaptive weight to obtain a denoised vibration signal.

8. The method according to claim 1, characterized in that Before inputting the standardized signal into the state recognition model, the method further includes: Get the initial weights of the initial state recognition model; Acquire multiple sample signals and a classification label corresponding to each of the sample signals; Inputting each of the sample signals and the corresponding classification labels into an initial state recognition model, training the initial state recognition model, and obtaining a loss function result of the initial state recognition model; wherein the loss function is determined based on the wavelet basis, the pseudo-inverse Jacobian stability term, and the time-frequency comparison term in the preprocessing model; According to the loss function result, the initial weight is updated to obtain the updated weight; Counting the number of updates of the updated weights, and returning to the step of inputting the sample signal and the corresponding classification label into an initial state recognition model, training the initial state recognition model, and obtaining a loss function result of the initial state recognition model, until the number of updates is greater than or equal to a preset number of updates; The initial state recognition model is set according to the updated weights to obtain a state recognition model.

9. The method according to claim 8, characterized in that The updating of the initial weights according to the loss function result to obtain updated weights includes: According to the loss function result, the initial weight is updated to obtain a first weight; For each sample signal, obtaining a sample lubrication state corresponding to the output of the initial state recognition model; For each sample signal, determining the fuzzy membership of the sample signal according to the sample lubrication state; The first weight is adjusted according to the fuzzy membership of each of the sample signals to obtain an updated weight.

10. An intelligent identification device for poor lubrication of equipment for equipment integrity management, characterized by: include: Signal acquisition module, used to obtain vibration signals of key rotating equipment in coal mines; a preprocessing module, configured to input the vibration signal into a preprocessing model, perform noise reduction and standardization on the vibration signal, and obtain a standardized signal; wherein the preprocessing model is configured to perform noise reduction on the vibration signal using an adaptive noise reduction strategy based on dynamic wavelet packet decomposition; An identification module is used to input the standardized signal into a state recognition model to obtain the lubrication status of the key rotating equipment in the coal mine; the state recognition model includes a multi-scale convolution module, a spatiotemporal attention weighting module, a feature fusion module and a classifier; the state recognition model is used to selectively enhance and fuse the multi-scale signal features of the standardized signal in the time domain and frequency domain, and classify based on the fused signal features to obtain the lubrication status of the key rotating equipment in the coal mine.

Images