Complex equipment fault prediction method and system based on multi-dimensional time sequence characteristic modulation

Abstract

The invention relates to the technical field of fault prediction, in particular to a complex equipment fault prediction method and system based on multi-dimensional time sequence characteristic modulation. The method comprises the following steps: preprocessing acquired sensor data; performing initial feature extraction by using the preprocessed sensor data; performing multi-dimensional time sequence feature coding and multi-scale feature modulation based on the extracted initial features; performing feature refining and time sequence smoothing on the modulated features; and obtaining a fault classification result and a maintenance suggestion. The multi-dimensional data isomerism is eliminated through the data preprocessing module, and the DTFE module extracts global features from multiple domains, so that the diagnosis precision, robustness and real-time performance are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault prediction, and in particular to a method and system for predicting complex equipment faults based on multi-dimensional time series feature modulation. Background Art

[0002] As an important part of key industrial systems, the operational stability of complex petroleum equipment is directly related to the efficiency and safety of the entire production system. During the long-term service of equipment, due to factors such as environmental changes, load impact, and component aging, the equipment may gradually evolve from minor abnormalities to serious failures. In order to ensure that the equipment maintains good performance during operation, there is an urgent need for a healthy operation diagnosis method that can monitor the equipment's operating status in real time, accurately identify potential risks, and predict possible failure trends. With the continuous development of industrial Internet of Things and artificial intelligence, more and more studies are trying to use time series data collected by multi-channel sensors to achieve equipment status assessment in a data-driven manner. However, existing diagnostic technologies still face many challenges, especially in processing equipment systems with complex coupling relationships and multi-source heterogeneous signals. Their recognition effect and generalization ability are far from meeting the requirements of engineering applications.

[0003] The current mainstream equipment health diagnosis methods mainly rely on traditional feature engineering or deep neural networks, which uniformly model the collected equipment signal data and then output the health status assessment results. However, in practical applications, the operating status of equipment often has obvious dynamic nonlinear and multi-scale evolution characteristics. Its signals contain both long-term trend evolution and short-term fluctuation anomalies, which makes it difficult for a single-scale, one-way modeling method to accurately capture key features. In addition, the interaction between different components may cause abnormal responses across channels, and the existing methods have major defects in modeling the dependency between the global and local states of the equipment. Furthermore, some signal anomalies initially manifest as small fluctuations or short-term mutations. Due to the limited perception ability of traditional models, it is often difficult to detect them in time, delaying the fault warning window and increasing maintenance risks. Therefore, there is an urgent need for an intelligent diagnosis method that integrates multi-dimensional structural information and time series evolution laws to comprehensively improve the operating status modeling ability and real-time health judgment ability of complex equipment.

[0004] To this end, the present invention proposes a complex equipment health operation diagnosis method and system based on multi-dimensional time series feature modulation. Based on fully exploring the internal laws of equipment operation data, this method combines multi-dimensional time series feature encoding and multi-dimensional time series feature modulation to effectively capture the coupling characteristics between global trends and local anomalies, and improves the stability and robustness of continuous state recognition through a time series smoothing mechanism, thereby achieving accurate diagnosis and early warning of the equipment operation health status. Summary of the invention

[0005] In order to solve the problems of the prior art in multi-dimensional time series feature extraction, weak fault signal detection and dynamic working condition adaptability, the present invention provides a complex equipment fault prediction method and system based on multi-dimensional time series feature modulation. Through the deep one-dimensional convolutional neural network combined with multi-dimensional time series feature encoding (DTFE), multi-dimensional time series feature modulation (MTFT), time series smoothing and fault classification technology, the diagnostic accuracy, robustness and real-time performance are comprehensively improved.

[0006] In the first aspect, the present invention provides a complex equipment fault prediction method based on multi-dimensional time series feature modulation, which adopts the following technical solution: A complex equipment fault prediction method based on multi-dimensional time series feature modulation, comprising: Get sensor data; Preprocess the acquired sensor data; Perform initial feature extraction using preprocessed sensor data; Multi-dimensional temporal feature encoding and multi-scale feature modulation are performed based on the extracted initial features; Perform feature refinement and time series smoothing on the modulated features; Get fault classification results and maintenance recommendations.

[0007] Furthermore, the preprocessing of the acquired sensor data includes standardizing the acquired sensor data, decomposing the signal using wavelet denoising and filtering out high-frequency noise; and then performing data enhancement by random time shift, which is expressed as: in, For denoised data; is a random time shift, which follows a uniform distribution ([-10,10]).

[0008] Furthermore, the initial feature extraction using the preprocessed sensor data includes performing multi-channel parallel convolution through a one-dimensional convolutional neural network to extract short-term time series features of multi-dimensional data; and then normalizing the distribution of the convolution output through layer normalization to process the multi-dimensional time series data to reduce internal covariate shift, which is expressed as: in, is the convolution output, k=1 is , k=2 is ; They are The mean and variance of A small constant used to prevent division by zero; are learnable parameters that control scaling and translation, respectively.

[0009] Furthermore, the use of preprocessed sensor data for initial feature extraction also includes using 1×1 convolution to integrate multi-sensor features through channel adjustment and weighted fusion to reduce redundancy and enhance the synergy of multi-dimensional data; in order to prevent complex equipment data from being affected by residual noise, resulting in model overfitting, adaptive maximum pooling is used to reduce the computational burden by compressing the time dimension, wherein Dropout enhances the robustness of the model by randomly discarding neurons, which is expressed as: in, is the fusion feature, and the dimension is ; For adaptive maximum pooling, the window size is (w=2), the stride is 2, and the output dimension is ; 30% of neurons are randomly dropped to prevent overfitting.

[0010] Furthermore, the multi-dimensional time series feature encoding and multi-scale feature modulation are performed based on the extracted initial features, including multi-dimensional feature pooling based on time domain pooling, frequency domain pooling and statistical domain pooling, and the obtained time domain, frequency domain and statistical domain features are formed into a multi-dimensional feature table through feature splicing; finally, redundant information is compressed by convolution to reduce the dimension, enhance the time series correlation, improve the feature quality and reduce the computational burden, which is expressed as: in, is the concatenation feature, and its dimension is ; for Convolution, dimension reduction to 128 channels; is the intermediate feature, dimension ; The convolution kernel size is ( k = 3 ), the number of output channels is ( C = 64 ), the stride is 1, and the padding is 1; is the convolution output, dimension ; Normalize the layers to prevent gradient explosion; is the activation function, introducing nonlinearity, Z is the final output, dimension .

[0011] Furthermore, the multi-dimensional time series feature encoding and multi-scale feature modulation based on the extracted initial features also include further extracting diversified time series patterns through multi-scale convolution, enhancing the perception of diversified fault modes, and obtaining features of different time scales; integrating multi-scale features to form a unified feature representation; finally, after enhancing the key frequency components through the frequency domain attention mechanism and highlighting the key sensor features through the channel attention mechanism, the computational complexity is reduced through feature dimensionality reduction to support real-time diagnosis, which is expressed as: in, is the channel attention output, dimension ; It is the DTFE output; It is a 1×1 convolution, which increases the dimension of Z to 128 channels; To fuse features, dimension ; It is a 1×1 convolution, reducing the dimension to 64 channels; Normalize the layer; is the activation function.

[0012] Furthermore, the modulated features are subjected to feature refinement and temporal smoothing, including enhancing high-order features through convolution refinement, and directly adding the MTFT output to the refined features based on the residual connection to ensure the original features, wherein the residual connection captures high-order temporal patterns through one-dimensional convolution to enhance the expressiveness of the features; and batch normalization is used to standardize the feature distribution, reduce internal covariate shift, and improve training stability and generalization ability; it is expressed as: in, It is a one-dimensional convolution with kernel size k=3, number of output channels (C=128), stride 1, and padding 1; is the mean and variance of the batch data, calculated by channel; A small constant to prevent division by zero; are learnable parameters that control scaling and translation, respectively; is the MTFT output, dimension ; The convolution is 1×1, increasing the number of channels from 64 to 128.

[0013] Furthermore, the feature refinement and time series smoothing of the modulated features also include compressing the time dimension through adaptive maximum pooling while retaining key time series features; and because the time series data of complex equipment presents non-stationary characteristics under dynamic conditions, the diagnosis results caused by feature mutations are unstable, and time series smoothing smoothes the dynamic changes of features through exponential moving average, thereby improving feature consistency and diagnostic reliability, which is expressed as: in, For adaptive maximum pooling, the window size is w=2 and the stride is 2; is the pooling feature at the current moment, dimension ; is the smooth feature of the previous moment, dimension ; is the smoothing factor, which controls the weight of new and old features.

[0014] Furthermore, the fault classification results and maintenance suggestions are obtained, including generating fault probabilities through a multi-layer fully connected network to ensure classification accuracy and robustness; and transforming them into specific suggestions through rule mapping to support decision-making in industrial scenarios to complete the diagnosis process, which is expressed as: in, is the first fully connected layer parameter; It is an intermediate feature with a dimension of 32; is the parameter of the second fully connected layer, and the 5 in the 5th power of the real matrix R is the number of fault categories (normal, wear, crack, overheating, electrical failure); To score the score, the dimension is 5; Normalization function to generate probability distribution; P is the failure probability and the dimension is 5.

[0015] In the second aspect, a complex equipment fault prediction system based on multi-dimensional time series feature modulation includes: A data acquisition module is configured to acquire sensor data; A preprocessing module is configured to preprocess the acquired sensor data; A feature extraction module is configured to perform initial feature extraction using the preprocessed sensor data; The modulation module is configured to perform multi-dimensional temporal feature encoding and multi-scale feature modulation based on the extracted initial features; The smoothing module is configured to perform feature refinement and time series smoothing on the modulated features; The classification module is configured to obtain fault classification results and maintenance suggestions.

[0016] In a third aspect, the present invention provides a computer-readable storage medium storing a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device, for example, a complex equipment fault prediction method based on multi-dimensional timing feature modulation.

[0017] In a fourth aspect, the present invention provides a terminal device comprising a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, wherein the instructions are suitable for being loaded by the processor and executing the complex equipment fault prediction method based on multi-dimensional timing feature modulation.

[0018] In summary, the present invention has the following beneficial technical effects: Compared with the defects of the prior art in single domain analysis, weak signal detection, dynamic working condition adaptability and coordination of global and local features, the present invention eliminates the heterogeneity of multidimensional data through the data preprocessing module, extracts global features from multiple domains through the DTFE module, enhances the perception of weak fault signals through the MTFT module, optimizes the feature stability under dynamic working conditions through the feature refinement and smoothing module, and accurately predicts faults and generates suggestions through the fault classification and suggestion module. In scenarios such as fracturing equipment and wind power gearboxes, the early fault missed diagnosis rate is reduced from 25% to 8%, the misdiagnosis rate is reduced from 15% to 6%, the feature extraction robustness is improved by 20%, and the single diagnosis time is shortened from 1.2 seconds to 0.5 seconds, which significantly improves the diagnostic accuracy, robustness and real-time performance.

[0019] The present invention realizes comprehensive analysis and processing of multi-dimensional data of complex equipment through multi-dimensional time series feature modulation technology. The DTFE module extracts global features from the time domain, frequency domain and statistical domain, and the MTFT module combines multi-scale convolution and attention mechanism to enhance weak signal perception. The multi-module collaborative design effectively responds to the challenges of dynamic working conditions, significantly improving the adaptability and reliability of complex equipment fault prediction and diagnosis, and meeting the needs of high-reliability scenarios.

[0020] This invention optimizes the application effect of predictive maintenance strategies through accurate fault prediction and diagnosis. In industrial scenarios such as fracturing equipment and wind power gearboxes, the technical solution supports efficient fault management, reduces the risk of equipment downtime, improves long-term reliability and operation and maintenance efficiency, and provides strong support for the development of industrial intelligence. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a schematic diagram of a complex equipment fault prediction method based on multi-dimensional time series feature modulation according to Example 1 of the present invention.

[0022] Figure 2 It is a diagram showing the missed diagnosis rate, misdiagnosis rate and diagnosis time effect of Example 1 of the present invention.

[0023] Figure 3 4 is a graph showing the accuracy and applicability of the suggestions of Example 1 of the present invention.

[0024] Figure 4 This is a radar chart of the comprehensive performance of Example 1 of the present invention.

[0025] Figure 5 It is a simulation diagram comparing the predicted advance amount and the actual fault occurrence time of Example 1 of the present invention. DETAILED DESCRIPTION

[0026] The present invention is further described in detail below in conjunction with the accompanying drawings.

[0027] Example 1 Reference Figure 1 , a complex equipment fault prediction method based on multi-dimensional time series feature modulation of this embodiment includes: Get sensor data; Preprocess the acquired sensor data; Perform initial feature extraction using preprocessed sensor data; Multi-dimensional temporal feature encoding and multi-scale feature modulation are performed based on the extracted initial features; Perform feature refinement and time series smoothing on the modulated features; Get fault classification results and maintenance suggestions; Specifically: (1). Multi-channel data preprocessing and feature initialization, During the operation of complex equipment, sensors will continuously collect multiple types of signals such as temperature, vibration, current, voltage, displacement, etc. Since signals from different channels have significant differences in amplitude, frequency and dimension, if they are directly input into the diagnostic model, it may cause feature suppression or response imbalance. Therefore, this step first converts the multi-source sensor data of complex equipment (such as vibration, temperature, current) into standardized multi-channel time series features, eliminates noise, enhances weak fault signals, and provides high-quality input for subsequent feature encoding.

[0028] 1) Data standardization, The sensor data of complex equipment (such as vibration, temperature, and current) have different dimensions and numerical ranges. For example, the vibration signal is expressed in acceleration (m / s²) and the temperature is expressed in degrees Celsius (℃). Direct processing will lead to an imbalance in feature weights and affect the model's ability to fuse multidimensional data. Data standardization eliminates dimensional differences by mapping each sensor data to a unified range, ensuring that subsequent convolution operations can fairly process multidimensional time series features. This is especially important for complex equipment such as fracturing equipment, because its multi-sensor data needs to be analyzed collaboratively to capture the operating status of multiple coupled components. The formula is as follows: in, is the original value of sensor m at time t, , T is the length of the time series; are the minimum and maximum values ​​of the data of sensor m.

[0029] 2) Wavelet denoising, Although data standardization unifies the dimensions, sensor data of complex equipment is still affected by noise. For example, the vibration signal of the wind turbine gearbox of the ultra-high pressure cutting system contains random noise due to wind speed changes, which masks the weak signal of early wear. Direct feature extraction of standardized data may cause noise to be mistaken for fault features, reducing diagnostic accuracy. By using wavelet denoising to decompose the signal and filter out high-frequency noise, the low-frequency and medium-frequency signals related to the fault are retained, and the detection capability of weak fault signals is enhanced. The formula is as follows:

[0030] in, is the standardized sensor data, ranging from [0,1]; For the wavelet denoising function, Daubechies wavelet (db4) is used; is the denoising threshold, which is used to control the noise filtering strength.

[0031] 3) Data augmentation, Since complex equipment is often in dynamic working conditions (such as load changes in oil-water separation equipment), the distribution of time series data changes over time. Although the denoised data is relatively clean, static data cannot fully simulate the diversity of dynamic working conditions, which may lead to insufficient generalization of the model in practical applications. Data enhancement simulates changes in working conditions through random time shifts, which will enhance the robustness of the model to dynamic environments. The formula is as follows: in, For denoised data; is a random time shift, which follows a uniform distribution ([-10,10]).

[0032] 4) Initial feature extraction, The operating status of complex equipment is jointly characterized by multi-dimensional sensor data. Although the denoised and enhanced data have high quality, it is still necessary to extract features that can reflect short-term time series patterns to capture the local characteristics of fault signals (such as the instantaneous impact of vibration signals). Through multi-channel parallel convolution of a one-dimensional convolutional neural network, short-term time series features of multi-dimensional data are extracted to provide a basis for subsequent global feature encoding.

[0033] in, , is the enhanced multi-sensor data, T=1024, M=3; dimensional convolution, convolution kernel size (k=3), number of output channels (C=32), stride of 1, and padding of 1.

[0034] 5) Layer normalization, The high dynamics of complex equipment data may lead to unstable training and affect model convergence. The distribution of convolution outputs will be standardized through layer normalization to process multi-dimensional time series data, reduce internal covariate shift, and improve training stability and generalization ability. The formula is as follows: in, is the convolution output, k=1 is , k=2 is ; They are The mean and variance of A small constant used to prevent division by zero; are learnable parameters that control scaling and translation, respectively.

[0035] 6) Feature fusion, Although parallel convolution extracts features of different depths, the operating status of complex equipment needs to be analyzed through multi-sensor data collaboration, which needs to be fused to generate a unified feature representation, reduce redundancy and enhance the synergy of multi-dimensional data. Here, 1×1 convolution is used to integrate multi-sensor features through channel adjustment and weighted fusion. The formula is as follows: in, is the normalized feature, and the dimensions are and ; for Convolution, which changes the number of channels from Adjusted to 64; for Convolution, which changes the number of channels from Adjusted to 64.

[0036] 7) Pooling and regularization, To prevent complex equipment data from being affected by residual noise, which may lead to model overfitting, adaptive maximum pooling is used to reduce the computational burden by compressing the time dimension (the original time dimension T=1024). Dropout enhances the robustness of the model by randomly discarding neurons. The formula is as follows: in, is the fusion feature, and the dimension is ; For adaptive maximum pooling, the window size is (w=2), the stride is 2, and the output dimension is ; 30% of neurons are randomly dropped to prevent overfitting.

[0037] (2). Multidimensional Temporal Feature Encoding (DTFE), The fault characteristics of complex equipment are distributed in multiple domains. For example, the vibration signal of the tensioner equipment is manifested as harmonic anomalies in the frequency domain, the temperature signal shows trend changes in the time domain, and the statistical domain reflects the non-normal distribution of the data. Traditional single-domain analysis (such as time domain only) ignores the diversity of features, resulting in insufficient modeling of the global operating state and difficulty in capturing weak fault signals. DTFE uses multi-dimensional pooling and feature fusion to integrate time domain, frequency domain and statistical domain features to extract the global timing features of complex equipment to capture the multi-dimensional characteristics of the operating state and provide a comprehensive feature representation for subsequent feature modulation.

[0038] 1) Time domain pooling, The time series data of complex equipment contains short-term trends and fluctuations. For example, the vibration signal of fracturing equipment reflects load changes in a short period of time. Direct processing of long time series data may ignore these local patterns and reduce diagnostic accuracy. Time domain pooling extracts local time series features by calculating short-term mean and standard deviation, captures the operating status under dynamic conditions, and provides a time domain basis for global feature encoding. The output (X) of step 1 contains multi-dimensional local features. It is necessary to extract short-term patterns through time domain pooling to construct a global representation: in, is the output of step 1, , ; is the pooling window size, with a stride of 2; is the channel index; Short-term mean of channel (c). is the short-time standard deviation of channel c.

[0039] Final Output , the number of channels is .

[0040] 2) Frequency domain pooling, In addition, the fault signals of complex equipment often appear as abnormalities of specific frequency components in the frequency domain. For example, the wear of wind turbine gearbox gears in oil-water separation equipment leads to enhanced high-frequency harmonics. Although time domain pooling extracts short-term trends, it cannot capture frequency domain features, which limits the ability to detect periodic faults. Frequency domain pooling is used to extract the main frequency components through fast Fourier transform (FFT), enhance the perception of periodic fault signals to capture periodic features, and improve the global feature representation. The formula is as follows: in, is the time series data of channel c, with dimension ; FFT is fast Fourier transform, which is used to extract frequency components; are the first 16 main frequency components; For frequency domain features, the time dimension is aligned by linear interpolation.

[0041] 3) Statistical domain pooling, Since the operating data of complex equipment often presents non-normal distribution, although the time domain and frequency domain features provide trend and periodicity information, they cannot capture the statistical characteristics of data distribution, limiting the comprehensive description of abnormal conditions. Statistical domain pooling calculates skewness and kurtosis to enhance the feature's ability to express non-normal distribution and improve multi-dimensional features. The formula is as follows: in, is the characteristic value of channel c at time t; is the mean and standard deviation of channel (c); is the skewness of channel c, reflecting the asymmetry of data distribution; is the kurtosis of channel (c), reflecting the sharpness of data distribution.

[0042] Final Output , the number of channels is .

[0043] 4) Feature stitching, The operating status of complex equipment needs to be described through multi-domain features, combining time domain trends, frequency domain harmonics and statistical distribution. Individual time domain, frequency domain or statistical domain features cannot fully characterize the fault mode, and need to be spliced ​​to form a multi-dimensional feature representation to improve the comprehensiveness of diagnosis. The formula is as follows: in, is the time domain feature; is the frequency domain feature; is the statistical domain feature; For the concatenated features, the number of channels is 128+16+128=272.

[0044] 5) Feature dimensionality reduction and enhancement, The real-time diagnosis of complex equipment requires more efficient feature representation. The number of multi-dimensional feature channels after splicing is large (272), which increases the computational complexity. There may be redundancy between features, and direct use for subsequent processing will reduce efficiency. By reducing the dimension through convolution compression of redundant information, the temporal correlation will be enhanced, the feature quality will be improved, and the computational burden will be significantly reduced. The formula is as follows: in, is the concatenation feature, and its dimension is ; for Convolution, dimension reduction to 128 channels; is the intermediate feature, dimension ; The convolution kernel size is ( k = 3 ), the number of output channels is ( C = 64 ), the stride is 1, and the padding is 1; is the convolution output, dimension ; Normalize the layers to prevent gradient explosion; is the activation function, introducing nonlinearity, Z is the final output, and the dimension .

[0045] (3). Multi-dimensional temporal feature modulation (MTFT), The fault characteristics of complex equipment (such as engine blade cracks) are manifested in different time scales (such as short-term vibration, long-term temperature drift) and frequency ranges. Traditional single-scale analysis is difficult to capture weak signals and has a high rate of missed diagnosis. MTFT captures diverse features through multi-scale convolution, and enhances the weight of key fault signals through frequency domain and channel attention mechanisms, solves the problem of weak signal detection, modulates multi-dimensional time series features, enhances the perception of weak fault signals, and optimizes feature expression.

[0046] 1) Multi-scale convolution, The fault signals of complex equipment have multi-time scale characteristics. For example, the short-term vibration shock of the wind turbine gearbox reflects instantaneous abnormalities, and the long-term temperature drift reflects accumulated wear. A single convolution kernel size cannot capture these features at the same time and may ignore weak fault signals. The output (X) of step 1 contains local features, and it is necessary to further extract diversified time series patterns through multi-scale convolution to enhance the perception of diversified fault modes and provide rich input for the subsequent attention mechanism. The formula is as follows: in, is the output of step 1; is the convolution kernel size, corresponding to the i-th branch; One-dimensional convolution, output channel number (C=32), stride 1, padding 1, 2, 3, 4 respectively; represents the output of the ith branch.

[0047] 2) Feature splicing, Multi-scale convolution generates features at different time scales, but using any scale alone cannot fully characterize the failure mode of complex equipment. For example, blade cracks in tensioner equipment require a combination of short-term and long-term feature analysis. Feature concatenation integrates multi-scale features to form a unified feature representation, providing comprehensive input for the subsequent attention mechanism. The formula is as follows: in, Represents the output of convolution at each scale, dimension ; is the concatenation feature, and the number of channels is .

[0048] 3) Frequency domain attention, The fault signals of complex equipment often appear as abnormalities of specific frequencies in the frequency domain. For example, electrical faults in fracturing equipment lead to enhancement of specific frequency components. Although multi-scale features contain rich time series information, they do not highlight key signals in the frequency domain, which may cause weak faults to be ignored. Frequency domain attention generates frequency weights through FFT and convolution, enhances the contribution of key frequency components, and improves the detection capability of weak fault signals. The formula is as follows: in, is a multi-scale feature, dimension ; is a fast Fourier transform that generates a frequency domain representation. is the frequency domain feature, and its dimension is ; It is a 1×1 convolution, reducing the dimension to 64 channels; is the intermediate feature, dimension ; It is a 1×1 convolution, restored to 128 channels; For the Sigmoid function, generate attention weights; is the frequency domain attention map, dimension ; is element-wise multiplication; is the modulated feature, dimension .

[0049] 4) Channel attention, Due to the different contributions of multi-dimensional sensor data of complex equipment, for example, the vibration signal of a wind turbine gearbox contributes more to fault detection than the temperature signal. Although frequency domain modulation enhances frequency characteristics, it does not optimize the weight distribution between channels, which may cause secondary features to interfere with diagnosis. Channel attention generates channel weights through global pooling and fully connected layers, highlights key sensor features, and improves the efficiency of multi-dimensional data fusion to highlight key channels. The formula is as follows: in, is the frequency domain modulation feature, dimension ; It is the result of global average pooling, with a dimension of 128; The parameters of the first fully connected layer are reduced to 64; It is an intermediate feature with a dimension of 64; is the parameter of the second fully connected layer, and the dimension is restored to 128; Channel attention map, dimension 128; is the channel modulation feature, dimension .

[0050] 5) Feature fusion and dimensionality reduction, The diagnosis of complex equipment needs to balance the global operating status and local fault characteristics. For example, the overall vibration mode of the fracturing equipment reflects the health level, and the local crack signal determines the diagnosis accuracy. Although the channel attention feature optimizes the local features, it needs to be integrated with the global features of DTFE to take both into account, and at the same time reduce the dimension to reduce the computational complexity and support real-time diagnosis. The formula is as follows: in, is the channel attention output, dimension ; It is the DTFE output; It is a 1×1 convolution, which increases the dimension of Z to 128 channels; To fuse features, dimension ; It is a 1×1 convolution, reducing the dimension to 64 channels; Normalize the layer; is the activation function.

[0051] (4). Feature refinement and time series smoothing, Complex equipment (such as fracturing equipment) generates non-stationary time series data under dynamic conditions. Traditional feature processing methods are susceptible to noise and mutation, resulting in unstable diagnosis. This step ensures feature stability through residual connection, extracts high-order features through convolution, smoothes dynamic changes through time series, optimizes the stability and consistency of multi-dimensional time series features, and adapts to the real-time diagnosis needs of complex equipment under dynamic conditions.

[0052] 1) Convolutional refinement, batch normalization and residual connection, Although the features of MTFT output have been optimized, the fault mode of complex equipment contains high-order time series features, which need to be further extracted to improve the diagnostic accuracy. Convolutional refinement captures high-order time series patterns through one-dimensional convolution to enhance the expressiveness of features; and uses batch normalization to standardize feature distribution, reduce internal covariate shift, and improve training stability and generalization ability; since the diagnosis of complex equipment needs to retain the original feature information, convolutional refinement enhances high-order features, but may lose some original information. Residual connection ensures the retention of original features by adding MTFT output directly to refined features, improving model stability and diagnostic accuracy. The formula is as follows: in, It is a one-dimensional convolution with kernel size (k=3), number of output channels (C=128), stride of 1, and padding of 1; is the mean and variance of the batch data, calculated by channel; A small constant to prevent division by zero; are learnable parameters that control scaling and translation, respectively; is the MTFT output, dimension ; The convolution is 1×1, increasing the number of channels from 64 to 128.

[0053] 2) Adaptive pooling and temporal smoothing, Real-time diagnosis of complex equipment requires efficient calculations. The time dimension of residual connection features is still high, which increases the computational complexity. Adaptive maximum pooling compresses the time dimension to reduce the computational burden while retaining key time series features. In addition, since the time series data of complex equipment exhibits non-stationary characteristics under dynamic conditions, feature mutations may lead to unstable diagnostic results. Time series smoothing uses exponential moving average to smooth the dynamic changes of features, improve feature consistency and diagnostic reliability, and the formula is as follows: in, For adaptive maximum pooling, the window size is w=2 and the stride is 2; is the pooling feature at the current moment, dimension ; is the smooth feature of the previous moment, dimension ; is the smoothing factor, which controls the weight of new and old features.

[0054] (5). Fault classification and maintenance suggestions, Complex equipment (such as fracturing equipment) has various types of faults (such as blade cracks and bearing wear), which require accurate classification to formulate maintenance strategies. Traditional classification methods are insensitive to weak fault signals and cannot provide real-time recommendations under dynamic conditions. This step maps the refined multi-dimensional time series features into fault probabilities through global pooling, multi-task pre-training, and rule mapping, generates maintenance recommendations for complex equipment, and supports accurate diagnosis and decision-making.

[0055] 1) Global average pooling and multi-task pre-training, The refined features contain time series dimensions, which will increase the computational complexity if used directly for classification. In addition, the diagnosis of complex equipment requires the overall operation status. Global average pooling compresses the time series dimension, extracts global features, reduces the computational burden, and generates a comprehensive representation. The sample size of fault data for complex equipment is limited, and direct training of the classifier may lead to overfitting and reduce generalization ability. Multi-task pre-training uses historical data to initialize the classifier, enhance the model's adaptability to small sample scenarios, and improve the ability to detect rare faults. The formula is as follows: in, is the output of step 4; It is the global pooling feature with a dimension of 128; are the pre-trained fully connected layer parameters.

[0056] 2) Fault classification and maintenance suggestion generation, Complex equipment has various types of faults, such as wear, cracks, and overheating of fracturing equipment systems, which need to be accurately distinguished by classifiers to support maintenance decisions. Although global pooling and pre-training optimize feature representation, it is necessary to generate fault probabilities through multi-layer fully connected networks to ensure classification accuracy and robustness; and the diagnosis of complex equipment not only needs to identify fault types, but also needs to provide actionable maintenance suggestions. Although the fault probability provides classification results, it needs to be converted into specific suggestions through rule mapping to support decision-making in industrial scenarios and complete the diagnosis process. The formula is as follows: in, is the first fully connected layer parameter; It is an intermediate feature with a dimension of 32; is the parameter of the second fully connected layer, and the 5 in the 5th power of the real matrix R is the number of fault categories (normal, wear, crack, overheating, electrical failure); To score the score, the dimension is 5; Normalization function to generate probability distribution; P is the failure probability and the dimension is 5.

[0057] Experimental verification: The experiment selected the operating data of the fracturing equipment, including multi-dimensional acoustic, optical, and electrical sensor data such as vibration and temperature, to construct a test data set including early faults and dynamic conditions. Construct a test data set including early faults and dynamic conditions. Set the load change rate (0, 0.2, 0.4, 0.6, 0.8, 1.0), where 0 indicates no load change and 1.0 indicates maximum load fluctuation. Compare the method of the present invention with five baseline methods: SVM (support vector machine), LSTM (long short-term memory network), CNN (convolutional neural network), RF (random forest) and DNN (deep neural network). Fault prediction and diagnosis are run on the same data set to evaluate the fault prediction accuracy, diagnostic stability under dynamic conditions, and the applicability of maintenance recommendations. The experimental results are as follows: Figure 2 , Figure 3 , Figure 4 As shown in Table 1, in order to display different data indicators in the same radar chart, the five indicators are processed relatedly: ① normalized to the interval [0,1]; ② reverse processing is performed on the "smaller the better" indicators (such as missed diagnosis rate, misdiagnosis rate, diagnosis time); ③ the "bigger the better" indicators (accuracy, suggestion applicability) are directly normalized. The method of the present invention is better than the baseline method at all load change rates.

[0058] Table 1 Data comparison of different methods under five indicators index SVM LSTM CNN RF DNN Method of the present invention Early failure missed diagnosis rate 26% 22% 28% 24% 20% 8% Misdiagnosis rate 16% 14% 18% 15% 12% 6% Single diagnosis time 1.3 seconds 1.5 seconds 1.1 seconds 1.4 seconds 1.2 seconds 0.5 seconds Fault prediction accuracy 78% 80% 76% 75% 82% 92% Maintenance Recommendations Applicability 72% 68% 70% 74% 78% 90% The experimental results show that SVM has strong classification ability, but limited processing of multidimensional data. LSTM is suitable for time series data, but weak in multidimensional data fusion. CNN is good at feature extraction, but lacks perception of weak signals. RF has good robustness, but lacks prediction of complex fault modes. DNN has strong performance, but is slightly inferior in processing data heterogeneity. The method of the present invention is superior to the baseline method in all indicators, reflecting the comprehensive performance improvement of the technical solution.

[0059] In order to intuitively compare the performance of the "method of the present invention" with other methods (SVM, LSTM, CNN, RF, DNN) in terms of advance time of fault prediction, the present invention provides a simulation diagram of "advance time of prediction vs. actual fault occurrence time", such as Figure 5 The difference between the predicted failure time of each method and the actual failure time (100 seconds) is clearly shown through the line graph, reference line and lead time annotation.

[0060] Through simulation graph analysis, the method of the present invention performs well in fault prediction, and its 20-second prediction lead time is significantly better than SVM, LSTM, CNN, RF and DNN (8-12 seconds lead time). This method not only improves the timeliness of prediction, but also provides more response time for fault prevention, and has high practical value and application potential.

[0061] Example 2 This embodiment provides a complex equipment fault prediction system based on multi-dimensional time series feature modulation, including: The data acquisition module is configured as follows: A computer-readable storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device, for a complex equipment fault prediction method based on multi-dimensional timing feature modulation.

[0062] A terminal device includes a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, wherein the instructions are suitable for being loaded and executed by the processor to perform a complex equipment fault prediction method based on multi-dimensional timing feature modulation.

[0063] The above are all preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A complex equipment fault prediction method based on multi-dimensional time series feature modulation, characterized in that: include: Get sensor data; Preprocess the acquired sensor data; Perform initial feature extraction using preprocessed sensor data; Multi-dimensional temporal feature encoding and multi-scale feature modulation are performed based on the extracted initial features; Perform feature refinement and time series smoothing on the modulated features; Get fault classification results and maintenance recommendations.

2. A complex equipment fault prediction method based on multi-dimensional time series feature modulation according to claim 1, characterized in that: The preprocessing of the acquired sensor data includes standardizing the acquired sensor data, decomposing the signal using wavelet denoising and filtering out high-frequency noise, and then performing data enhancement by random time shift, which is expressed as: in, For denoised data; is a random time shift that follows a uniform distribution, i.e., [-10,10].

3. A complex equipment fault prediction method based on multi-dimensional time series feature modulation according to claim 2, characterized in that: The initial feature extraction is performed using the preprocessed sensor data, including performing multi-channel parallel convolution through a one-dimensional convolutional neural network to extract short-term time series features of the multi-dimensional data; Then, the distribution of the convolution output is standardized by layer normalization to process multi-dimensional time series data to reduce internal covariate shift, which is expressed as: in, is the convolution output, k=1 is , k=2 is ; They are The mean and variance of A small constant used to prevent division by zero; are learnable parameters that control scaling and translation, respectively.

4. The method for predicting complex equipment faults based on multi-dimensional time series feature modulation according to claim 3 is characterized in that: The initial feature extraction using the preprocessed sensor data also includes using 1×1 convolution to integrate multi-sensor features through channel adjustment and weighted fusion to reduce redundancy and enhance the synergy of multi-dimensional data; in order to prevent complex equipment data from being affected by residual noise, resulting in model overfitting, adaptive maximum pooling is used to reduce the computational burden by compressing the time dimension, wherein Dropout enhances the robustness of the model by randomly discarding neurons, which is expressed as: in, is the fusion feature, and the dimension is ; For adaptive maximum pooling, the window size is w=2, the stride is 2, and the output dimension is ; 30% of neurons are randomly dropped to prevent overfitting.

5. The method for complex equipment fault prediction based on multi-dimensional time series feature modulation according to claim 4 is characterized in that: The multi-dimensional time series feature encoding and multi-scale feature modulation based on the extracted initial features include multi-dimensional feature pooling based on time domain pooling, frequency domain pooling and statistical domain pooling, and the obtained time domain, frequency domain and statistical domain features are formed into a multi-dimensional feature table through feature splicing; finally, redundant information is compressed by convolution to reduce the dimension, enhance the time series correlation, improve the feature quality and reduce the computational burden, which is expressed as: in, is the concatenation feature, and its dimension is ; for Convolution, dimension reduction to 128 channels; is the intermediate feature, dimension ; The convolution kernel size is k=3, the number of output channels is C=64, the stride is 1, and the padding is 1; is the convolution output, dimension ; Normalize the layers to prevent gradient explosion; is the activation function, introducing nonlinearity, Z is the final output, and the dimension .

6. The method for complex equipment fault prediction based on multi-dimensional time series feature modulation according to claim 5, characterized in that: The multi-dimensional time series feature encoding and multi-scale feature modulation based on the extracted initial features also include further extracting diversified time series patterns through multi-scale convolution, enhancing the perception of diversified fault modes, and obtaining features of different time scales; integrating multi-scale features to form a unified feature representation; finally, after enhancing key frequency components through the frequency domain attention mechanism and highlighting key sensor features through the channel attention mechanism, the computational complexity is reduced through feature dimensionality reduction to support real-time diagnosis, which is expressed as: in, is the channel attention output, dimension ; It is the DTFE output; It is a 1×1 convolution, which increases the dimension of Z to 128 channels; To fuse features, dimension ; It is a 1×1 convolution, reducing the dimension to 64 channels; Normalize the layer; is the activation function.

7. The method for complex equipment fault prediction based on multi-dimensional time series feature modulation according to claim 6 is characterized in that: The modulated features are refined and time-series smoothed, including enhancing high-order features through convolutional refinement, and directly adding the MTFT output to the refined features based on the residual connection to ensure the original features, wherein the residual connection captures high-order time series patterns through one-dimensional convolution to enhance the expressiveness of the features; and batch normalization is used to standardize the feature distribution, reduce internal covariate shift, and improve training stability and generalization ability; it is expressed as: in, It is a one-dimensional convolution with kernel size k=3, output channel number C=128, stride 1, and padding 1; is the mean and variance of the batch data, calculated by channel; A small constant to prevent division by zero; are learnable parameters that control scaling and translation, respectively; is the MTFT output, dimension ; The convolution is 1×1, increasing the number of channels from 64 to 128.

8. The method for complex equipment fault prediction based on multi-dimensional time series feature modulation according to claim 7, characterized in that: The feature refinement and time series smoothing of the modulated features also include compressing the time dimension through adaptive maximum pooling while retaining key time series features; and because the time series data of complex equipment presents non-stationary characteristics under dynamic conditions, the diagnosis results caused by feature mutations are unstable, and time series smoothing uses exponential moving average to smooth the dynamic changes of features and improve feature consistency and diagnostic reliability, which is expressed as: in, For adaptive maximum pooling, the window size is w=2 and the stride is 2; is the pooling feature at the current moment, dimension ; is the smooth feature of the previous moment, dimension ; is the smoothing factor, which controls the weight of new and old features.

9. The method for complex equipment fault prediction based on multi-dimensional time series feature modulation according to claim 8, characterized in that: The fault classification results and maintenance suggestions are obtained, including generating fault probabilities through a multi-layer fully connected network to ensure classification accuracy and robustness; and transforming them into specific suggestions through rule mapping to support decision-making in industrial scenarios to complete the diagnosis process, which is expressed as: in, is the first fully connected layer parameter; It is an intermediate feature with a dimension of 32; is the parameter of the second fully connected layer, and the fifth power of the real matrix R represents the number of fault categories, namely normal, wear, crack, overheating, and electrical fault; To score the score, the dimension is 5; Normalization function to generate probability distribution; P is the failure probability and the dimension is 5.

10. A complex equipment fault prediction system based on multi-dimensional time series feature modulation, characterized in that: include: A data acquisition module is configured to acquire sensor data; A preprocessing module is configured to preprocess the acquired sensor data; A feature extraction module is configured to perform initial feature extraction using the preprocessed sensor data; The modulation module is configured to perform multi-dimensional temporal feature encoding and multi-scale feature modulation based on the extracted initial features; The smoothing module is configured to perform feature refinement and time series smoothing on the modulated features; The classification module is configured to obtain fault classification results and maintenance suggestions.

Images