Equipment early warning method based on intelligent diagnosis, equipment and medium

Abstract

The invention discloses an equipment early warning method and equipment based on intelligent diagnosis and a medium, and belongs to the technical field of equipment monitoring and early warning. The method comprises the following steps: deploying a temperature sensor network to a to-be-detected device to obtain original temperature time sequence data; normalizing the original temperature time sequence data to generate a normalized temperature value; determining a weighted second order gradient entropy feature based on the normalized temperature value; constructing an end-to-end fault diagnosis model; updating the fault prototype vector weight according to the equipment working condition change of the to-be-detected equipment; determining a confidence score; inputting temperature data collected in real time into the trained end-to-end fault diagnosis model to obtain fault probability distribution; and when the confidence score exceeds a confidence threshold and the fault probability is higher than a preset probability threshold, early warning is performed. According to the method, the technical effects of feature retention of non-stationary temperature data, cooperative detection of composite faults, multi-scale modeling and optimization of dynamic adaptability are achieved.

Description

Technical Field

[0001] The present application relates to the technical field of equipment monitoring and early warning, and in particular to an equipment early warning method, equipment, and medium based on intelligent diagnosis. Background Art

[0002] In the field of predictive maintenance technology for equipment, particularly methods based on intelligent diagnosis, existing technologies primarily involve deploying high-precision temperature sensor networks to key components of industrial equipment. These sensors collect raw temperature time series data in real time at a fixed sampling frequency, and transmit this data via an industrial IoT gateway to a central processing unit for preprocessing. Existing solutions include using conventional normalization methods such as min-max or Z-score to standardize temperature data to eliminate dimensional differences, while employing differential operations to extract gradient features to identify fault modes. In addition, machine learning models such as long short-term memory networks (LSTMs) are widely used in time series modeling to capture long-term dependencies in temperature data, while attention mechanisms are introduced to extract multi-scale features. These information is then integrated through a feature fusion layer, ultimately outputting a fault probability distribution for maintenance decision-making. These technologies have been applied to key industrial scenarios such as manufacturing and power generation, aiming to achieve early warning of equipment status and fault prediction through a data-driven approach, thereby reducing downtime and maintenance costs. The overall technology field is rapidly evolving, relying on the development of the Internet of Things, artificial intelligence, and big data processing, providing fundamental support for industrial automation.

[0003] Existing technologies for predictive maintenance of equipment still have significant shortcomings. First, conventional normalization methods ignore the temporal structure of temperature signals, resulting in the compression or loss of key features in areas of load fluctuation or mutation, and are unable to effectively retain mutation information under non-stationary operating conditions. Second, differential operations are sensitive to high-frequency noise or have a delayed response to slowly changing trends, making it difficult to simultaneously capture the significant patterns of complex faults such as bearing wear and cooling shock in the same framework, reducing the comprehensiveness and reliability of diagnosis. Third, existing models mostly use fixed windows or single LSTM structures, failing to consider temperature variation patterns at different scales. This leads to poor recognition of early anomalies and small fluctuation precursors, reducing early warning capabilities. Finally, most existing methods use static model parameters and lack dynamic adaptation mechanisms. Once the equipment operating conditions change (such as load configuration changes or ambient temperature drift), performance degrades rapidly, making long-term tracking and robust diagnosis impossible. These shortcomings ultimately lead to high false alarm rates, low detection accuracy, and poor long-term stability, limiting the application of predictive maintenance in complex industrial environments.

[0004] Therefore, how to achieve feature retention of non-stationary temperature data, collaborative detection of complex faults, multi-scale modeling, and optimization of dynamic adaptability has become a technical problem that needs to be solved urgently. Summary of the Invention

[0005] The embodiments of the present application provide an equipment early warning method, equipment, and medium based on intelligent diagnosis to solve the following technical problems: how to achieve feature retention of non-stationary temperature data, collaborative detection of complex faults, multi-scale modeling, and optimization of dynamic adaptability.

[0006] In the first aspect, the embodiment of the present application provides an equipment early warning method based on intelligent diagnosis, which includes: deploying a high-precision temperature sensor network to the key components of the equipment to be detected to obtain the original temperature time series data of the key components, and sending it to the preset central processing unit through the preset industrial Internet of Things gateway; normalizing the original temperature time series data through the central processing unit to generate a normalized temperature value; determining the weighted second-order gradient entropy feature based on the normalized temperature value; constructing an end-to-end fault diagnosis model; wherein the input of the end-to-end fault diagnosis model is the normalized temperature value and the weighted second-order gradient entropy feature, and the long-term memory network layer of the end-to-end fault diagnosis model is used to capture the long-term memory feature. The multi-scale fault features are extracted through the multi-head temporal attention algorithm based on the period dependency relationship. The multi-scale fault features are output as integrated features through the adaptive feature fusion layer, and the integrated features are output as fault probability distribution through the fully connected classification layer. The weight of the fault prototype vector is updated according to the equipment operating condition changes of the equipment to be detected. The confidence score is determined according to the feature deviation, classification uncertainty and gradient entropy change trend in the end-to-end fault diagnosis model. The real-time collected temperature data is sequentially generated into feature vectors and input into the trained end-to-end fault diagnosis model to obtain the fault probability distribution. When the confidence score exceeds the confidence threshold and the fault probability in the fault probability distribution continues to be higher than the preset probability threshold, an early warning is issued.

[0007] In one implementation of the present application, the original temperature time series data is normalized by a central processing unit to generate a normalized temperature value, specifically including: setting a time window length parameter, and dividing continuous overlapping time windows based on the time window length parameter; wherein the time window includes multiple data points, and the data points are temperature data corresponding to the time series in the original temperature time series; determining the sliding mean of the time window based on the arithmetic mean of all data points in the time window; determining the sliding standard deviation of the data points in the time window by the sum of the squares of the deviations between the data points in the window and the sliding mean; and calculating the temperature data of each time series in the time window based on the sliding mean and the sliding standard deviation to generate a normalized temperature value.

[0008] In one implementation of the present application, a weighted second-order gradient entropy feature is determined based on the normalized temperature value, specifically including: performing a differential operation based on the temperature values ​​of any three adjacent time points in the normalized temperature value to determine a second-order difference; wherein the second-order difference is used to suppress high-order noise; the calculation of the second-order difference is expressed by the following formula:

[0009] in, is the second-order difference, for The normalized temperature value at the moment, for The normalized temperature value at the moment, for Normalized temperature value at the moment; A fixed-length window is set. Within the fixed-length time window, the second-order difference is divided into multiple gradient value bins according to a preset range to determine the probability distribution of each gradient value bin; the bin weight coefficient is determined based on the mean and maximum gradient of the gradient value bins; and the weighted second-order gradient entropy feature is determined based on the probability distribution and the bin weight coefficient.

[0010] In one implementation of the present application, the construction of an improved long short-term memory network layer specifically includes: calculating the average temperature gradient value of the normalized temperature value within a sliding window containing the current time point; multiplying the absolute value of the average temperature gradient value by a preset gate enhancement coefficient to generate a gradient gate term; adding the gradient gate term to the original input term of the forget gate to determine the forget gate input; applying a sigmoid activation function to the enhanced forget gate input to determine the forget gate output; wherein the forget gate output determines the extent to which the cell state at the previous moment is retained.

[0011] In one implementation of the present application, multi-scale fault features are extracted through a multi-head temporal attention algorithm, specifically including: setting multiple independent attention heads; setting different time window scales for multiple independent attention heads; wherein the time window scale is the length of the historical time step range that the attention mechanism focuses on; generating a query vector based on the hidden state output of the long short-term memory network layer at the current moment; generating a key vector matrix and a value vector matrix based on the historical hidden state output sequence within the time scale window; calculating the dot product of the query vector and the key vector matrix, dividing it by a preset scaling factor, performing masking processing, and applying a softmax function to obtain a weight distribution; performing weighted summation on the value vector matrix based on the weight distribution to determine the output vector of the independent attention head at the current time window scale; splicing the output vectors of multiple independent attention heads to output multi-scale fault features.

[0012] In one implementation of the present application, multi-scale fault features are integrated through the output of an adaptive feature fusion layer, specifically including: splicing the current hidden state output of the long short-term memory network with the feature vector output by the multi-head temporal attention algorithm to obtain a spliced ​​feature vector; processing the spliced ​​feature vector based on the sigmoid activation function to generate a fused gated value vector; determining the weighted hidden state feature based on the hidden state output of the long short-term memory network layer and the fused gated value vector; determining the weighted attention feature based on the feature vector output by the multi-head temporal attention algorithm and the fused gated value vector; generating an integrated feature based on the weighted hidden state feature and the weighted attention feature.

[0013] In one implementation of the present application, updating the fault prototype vector weight according to the change in the device operating condition of the device to be detected specifically includes: based on Determine the constrained inter-class distance to determine the prototype contrast loss; where, For the class prototype; is the prototype contrast loss, is the number of fault categories, For the The mean of the class sample features, For category The characteristic output of the sample, is the boundary margin, For category The characteristic output of the sample, For category The characteristic output of the sample, is the Euclidean distance, For category The number of samples, is the sample set of category i, is the sample index, is the feature extraction network, is the nth sample, represents the maximum value function; A preset cross-entropy loss function is introduced; the total loss function is determined based on the prototype contrast loss and the cross-entropy loss function: the new batch data and historical batch data of the equipment to be tested are determined; when the batch difference between the new batch feature distribution of the new batch data and the historical feature distribution of the historical batch data is greater than the preset difference threshold, the update rate is determined according to the value of the batch difference; the fault prototype vector weight is updated based on the update rate.

[0014] In one implementation of the present application, a confidence score is determined based on the feature deviation, classification uncertainty and gradient entropy change trend in the end-to-end fault diagnosis model, specifically including: measuring the Euclidean distance between the current fusion feature vector and the feature prototype vector of the end-to-end fault diagnosis model, and dividing it by the maximum norm of all category feature prototype vectors to determine the feature deviation; calculating the fault probability distribution output by the end-to-end fault diagnosis model to determine the information entropy to determine the classification uncertainty; wherein, the higher the entropy value of the classification uncertainty, the greater the uncertainty; determining the gradient entropy change based on the absolute value of the difference between the weighted second-order gradient entropy eigenvalue at the current moment and the second-order gradient entropy eigenvalue at the previous moment; multiplying the feature deviation, classification uncertainty and gradient entropy change by preset weight coefficients and adding them together to determine a weighted comprehensive value; inputting the weighted comprehensive value into the sigmoid function for nonlinear mapping to determine the confidence score.

[0015] In a second aspect, an embodiment of the present application further provides an equipment early warning device based on intelligent diagnosis, the device comprising: at least one processor; and a memory in communication with the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so as to enable the at least one processor to: deploy a high-precision temperature sensor network to the key components of the equipment to be detected to obtain the original temperature time series data of the key components, and send it to the preset central processing unit through the preset industrial Internet of Things gateway; normalize the original temperature time series data through the central processing unit to generate a normalized temperature value; determine the weighted second-order gradient entropy feature based on the normalized temperature value; construct an end-to-end fault diagnosis model; wherein the input of the end-to-end fault diagnosis model is the normalized temperature value. The method combines the degree value and weighted second-order gradient entropy features, captures long-term dependencies through the improved long short-term memory network layer of the end-to-end fault diagnosis model, extracts multi-scale fault features through the multi-head temporal attention algorithm, outputs integrated features through the adaptive feature fusion layer, and outputs the fault probability distribution through the fully connected classification layer; updates the weight of the fault prototype vector according to the change of the equipment working condition of the equipment to be tested; determines the confidence score according to the feature deviation, classification uncertainty and gradient entropy change trend in the end-to-end fault diagnosis model; generates feature vectors from the real-time collected temperature data in sequence, and inputs them into the trained end-to-end fault diagnosis model to obtain the fault probability distribution; and issues an early warning when the confidence score exceeds the confidence threshold and the fault probability in the fault probability distribution continues to be higher than the preset probability threshold.

[0016] On the third aspect, the embodiment of the present application also provides a non-volatile determination machine storage medium for equipment early warning based on intelligent diagnosis, which stores determination machine executable instructions, and the determination machine executable instructions are set to: deploy a high-precision temperature sensor network to the key components of the equipment to be detected to obtain the original temperature time series data of the key components, and send it to the preset central processing unit through the preset industrial Internet of Things gateway; normalize the original temperature time series data through the central processing unit to generate a normalized temperature value; determine the weighted second-order gradient entropy feature based on the normalized temperature value; construct an end-to-end fault diagnosis model; wherein the input of the end-to-end fault diagnosis model is the normalized temperature value and the weighted second-order gradient entropy feature, and the end-to-end fault diagnosis model is used. The improved long short-term memory network layer captures long-term dependencies, extracts multi-scale fault features through the multi-head temporal attention algorithm, and the multi-scale fault features are output as integrated features through the adaptive feature fusion layer. The integrated features are output as fault probability distribution through the fully connected classification layer; the fault prototype vector weight is updated according to the equipment operating condition changes of the equipment to be tested; the confidence score is determined according to the feature deviation, classification uncertainty and gradient entropy change trend in the end-to-end fault diagnosis model; the real-time collected temperature data is sequentially generated into feature vectors and input into the trained end-to-end fault diagnosis model to obtain the fault probability distribution; an early warning is issued when the confidence score exceeds the confidence threshold and the fault probability in the fault probability distribution continues to be higher than the preset probability threshold.

[0017] The embodiments of the present application provide a device early warning method, device, and medium based on intelligent diagnosis, which have at least the following technical effects: In terms of data preprocessing, the piecewise adaptive normalization method effectively improves the feature retention of non-stationary temperature data. By dividing the dynamic time window, calculating the sliding mean and standard deviation, and performing local normalization at each time point, this method solves the problem of conventional min-max or Z-score normalization causing key feature loss in areas of load fluctuation or mutation.

[0018] During the feature extraction phase, weighted second-order gradient entropy features significantly enhance the detection sensitivity and noise immunity of complex faults. By calculating second-order differences to suppress high-frequency noise and combining them with weighted binning statistics of gradient distributions, this method amplifies the entropy changes for both slowly varying faults (such as bearing wear) and sudden faults (such as cooling shock). This method exhibits a continuously enhanced response for slowly varying faults and produces pure pulse peaks for transient faults, addressing the issues of response lag and noise sensitivity associated with traditional differential operations, thereby improving the comprehensiveness and stability of diagnosis.

[0019] In terms of model architecture design, the end-to-end fault diagnosis model significantly enhances multi-scale feature fusion capabilities and discrimination accuracy through improvements to the long-short-term memory network, a multi-head temporal attention mechanism, and an adaptive feature fusion layer. An improved LSTM forget gate combined with temperature gradient gating enhances long-term memory for slowly varying faults. The multi-head attention mechanism concurrently extracts fault features from different time windows, addressing the limitations of single-scale modeling. The adaptive feature fusion layer dynamically weights features based on the variance of temperature fluctuations, improving the model's adaptability to complex fault modes.

[0020] To address device state drift, the online dynamic weight update mechanism significantly enhances the long-term robustness of the model. This incremental prototype adjustment strategy adaptively updates the prototype vector weights when the feature distribution of a new batch differs significantly from the historical distribution. This addresses the performance degradation of static models under varying operating conditions, such as load configuration changes or ambient temperature drift.

[0021] In the fault warning phase, a multi-metric confidence quantification method significantly improves alarm accuracy and reliability. By jointly evaluating feature deviation, classification uncertainty, and gradient entropy trends to generate a comprehensive confidence score, this method addresses the high false alarm rate of traditional threshold methods. This method ensures that when the confidence level exceeds the threshold and the fault probability remains high, the triggered alarm is highly credible, thereby optimizing the practicality of maintenance decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 A flow chart of an equipment early warning method based on intelligent diagnosis provided in an embodiment of the present application; Figure 2 Schematic diagram of the original temperature signal; Figure 3 This is a comparison chart of the effects of different normalization methods; Figure 4 for Figure 3 Enlarged view of the medium-Gaussian fault area; Figure 5 A comparison chart of different feature extraction methods and the feature extraction method of this application; Figure 6 This is a comparison chart of the ablation experiment performance of different model architectures; Figure 7 This is a comparison chart of the performance changes of different prediction methods in long-term operation; Figure 8 A schematic diagram of the internal structure of an equipment warning device based on intelligent diagnosis provided in an embodiment of the present application. DETAILED DESCRIPTION

[0023] To make the purpose, technical solutions, and advantages of this application more clear, the technical solutions of this application will be clearly and completely described below in conjunction with the specific embodiments of this application and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0024] The embodiments of the present application provide an equipment early warning method, equipment, and medium based on intelligent diagnosis to solve the following technical problems: how to achieve feature retention of non-stationary temperature data, collaborative detection of complex faults, multi-scale modeling, and optimization of dynamic adaptability.

[0025] The technical solutions proposed in the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0026] Figure 1 This is a device warning flow chart based on intelligent diagnosis provided in the embodiment of this application. Figure 1 As shown, an embodiment of the present application provides an equipment early warning method based on intelligent diagnosis, which specifically includes the following steps: Step 1: Deploy a high-precision temperature sensor network to the key components of the equipment to be tested to obtain the original temperature time series data of the key components and send it to the preset central processing unit through the preset industrial Internet of Things gateway.

[0027] A "high-precision temperature sensor network" refers to a set of temperature sensors distributed across critical equipment components (such as bearings, motors, or cooling systems). These sensors feature high resolution (for example, ±0.1°C accuracy) and a fixed sampling frequency (for example, 1Hz) to continuously monitor temperature changes. "Critical components" are core components within the equipment that are prone to failure or could affect overall performance. Data collection must cover the full operating range (including startup, steady-state operation, load fluctuations, and shutdown). The "Industrial IoT Gateway" is a data transfer device responsible for transmitting raw temperature time-series data collected by sensors to the central processing unit in real time, ensuring data synchronization and integrity. The "pre-installed central processing unit" refers to a remote or local computing server for subsequent data processing.

[0028] In a specific case, equipment A required predictive maintenance. The maintenance team deployed three high-precision temperature sensors (sampling frequency 1Hz) at the reactor's agitator shaft seal to collect real-time temperature changes during operation. These sensors synchronously transmitted data to the central processing unit via the factory's Industrial IoT gateway. The data collection process covered the equipment's full operating range, including the reactor startup phase (temperature rising from room temperature to 150°C), steady-state operation (temperature fluctuations of ±2°C), sudden load changes (simulating sudden temperature changes caused by feed changes), and shutdown and cooling. This process generated a temperature time series dataset encompassing both normal conditions and potential seal failure modes.

[0029] Step 2: Normalize the original temperature time series data by the central processing unit to generate a normalized temperature value.

[0030] Step 2.1, set the time window length parameter, and divide the continuous overlapping time windows based on the time window length parameter; wherein the time window includes multiple data points, and the data points are the temperature data corresponding to the time series in the original temperature time series.

[0031] First, determine the time window length τ and half window width w to divide multiple consecutive overlapping time windows, and use the formula Determine the time window number k to which the current time point t belongs.

[0032] Step 2.2: Determine the sliding mean of the time window based on the arithmetic mean of all data points in the time window.

[0033] Furthermore, within the time window number k, by formula Determine the sliding mean; where, is the sliding mean, is the time index, for The original temperature value at the moment, is the total number of data points in the window, .

[0034] Step 2.3: Determine the sliding standard deviation of the data points in the time window by the sum of the squares of the deviations between the data points in the window and the sliding mean.

[0035] Furthermore, through the formula Determine the sliding standard deviation; where, is the sliding standard deviation, is the original temperature value at time t.

[0036] Step 2.4: Calculate the temperature data of each time series within the time window based on the sliding mean and sliding standard deviation to generate a normalized temperature value.

[0037] Furthermore, according to the formula Generates normalized temperature values; where γ k is the scaling factor, β k is the offset factor, ε is the smoothing factor, is the normalized temperature value.

[0038] In a specific example, the central processing unit performs piecewise adaptive normalization on the temperature monitoring data of device A (including non-stationary characteristics caused by load fluctuations). The time window length τ = 60 seconds (window length) and the half-window width w = 10 seconds (window radius) are set. For the data point at time t = 300 seconds, the sliding mean (the average of all temperature values ​​within the window for 300 ± 10 seconds) and the sliding standard deviation are calculated for the data point within the window (k = 300 / 60 = 5).

[0039] like Figures 2 to 4 As shown, in order to verify the feature retention ability of the segmented adaptive normalization method under complex working conditions, the method of the present invention is compared with the minimum-maximum normalization ( Figure 3 Min-Max normalization), Z score normalization ( Figure 3 In this paper, the three normalization methods are compared with Z-Score normalization to analyze the feature preservation effects of the three normalization methods in non-stationary temperature signals (including load fluctuations, instantaneous spikes, Gaussian faults and exponential rising faults).

[0040] like Figure 2 As shown, the raw temperature signal shows that the device has experienced multiple load changes and three typical fault events.

[0041] Traditional normalization methods (minimum-maximum value, Z score) show feature smoothing in the fault area, such as Figure 4 As shown, the key mutation characteristics are significantly weakened, especially in the Gaussian fault region (enlarged image at 2000-2100 points). The method of the present invention fully preserves the detailed features of all fault events: the steep rising edge of the instantaneous spike, the complete waveform of the Gaussian fault, and the gradual trend of the exponential rise are all clearly visible.

[0042] Experimental results show that piecewise adaptive normalization effectively solves the problem of feature loss in non-stationary signal processing through dynamic time window statistics.

[0043] Step 3: Determine the weighted second-order gradient entropy feature based on the normalized temperature value.

[0044] Step 3.1: Perform a differential operation based on the temperature values ​​of any three adjacent time points in the normalized temperature value to determine a second-order difference; wherein the second-order difference is used to suppress high-order noise.

[0045] The calculation of the second-order difference is expressed by the following formula:

[0046] in, is the second-order difference, for The normalized temperature value at the moment, for The normalized temperature value at the moment, for Normalized temperature value at the moment; Step 3.2: Set a fixed-length window. Within the fixed-length time window, divide the second-order difference into multiple gradient value bins according to a preset range to determine the probability distribution of each gradient value bin.

[0047] Furthermore, a gradient value bin set B is preset, and the probability of the second-order difference value falling into each bin b is calculated based on the gradient value bin set B within a time window of length L. , to construct gradient bins; where, Before and after The gradient value in the time window falls into the box The probability of is the counting function, is the time window length, The bin index.

[0048] Step 3.3: Determine the bin weight coefficient based on the mean and maximum gradient of the gradient value bins.

[0049] The box weight coefficient is expressed by the following formula:

[0050] in, For the The mean gradient within each box, is the maximum absolute value of the second-order gradient in the current time window.

[0051] Step 3.4: Determine the weighted second-order gradient entropy feature based on the probability distribution and the box weight coefficient.

[0052] Furthermore, through the formula Output weighted second-order gradient entropy features; where, is the weighted second-order gradient entropy.

[0053] In a specific example, for the temperature monitoring data of equipment A, a weighted second-order gradient entropy feature is first extracted. The second-order difference of the normalized temperature series is calculated (the value at time t = (temperature at time t+1 - 2 × temperature at time t + temperature at time t-1)) to suppress high-frequency noise. A set of gradient bins B (-0.5 to 0.5°C / second, with each bin consisting of 0.1°C / second intervals) is preset. The probability of each bin (e.g., the percentage of samples falling within the -0.1 to 0°C / second interval) is calculated within a time window of L = 30 seconds. Finally, the entropy feature is output based on the bin weights (positively correlated with the average absolute value of the gradient within the bin) to enhance sensitivity to bearing wear (slow-onset faults) and cooling failure (sudden faults).

[0054] like Figure 5 As shown, the fault sensitivity comparison of feature extraction methods is analyzed, validating the sensitivity of weighted second-order gradient entropy to complex fault detection. Comparing the response characteristics of the feature extraction method of the present invention with those of first-order and second-order gradient features in the combined scenarios of bearing wear (slow-changing fault) and cooling system shock (transient fault) reveals that the first-order gradient feature is sensitive to cooling shock but suffers from severe noise interference, with almost no response in the bearing wear region. While the second-order gradient feature suppresses noise, its response to slow-changing faults is delayed and insufficient. The method of the present invention demonstrates optimal characteristics for both fault types, exhibiting a continuously increasing, gradual response to bearing wear (yellow area after 100 seconds) and producing a significant and pure pulse peak to three cooling shocks (red area). A stable baseline is maintained in non-fault regions, demonstrating significant noise suppression. This feature achieves collaborative detection of slow-changing and sudden-changing faults through the fusion of entropy weighting and gradient distribution statistics.

[0055] Step 4. Construct an end-to-end fault diagnosis model; wherein, the input of the end-to-end fault diagnosis model is the normalized temperature value and the weighted second-order gradient entropy feature, the long-term dependency is captured by the improved long short-term memory network layer of the end-to-end fault diagnosis model, and the multi-scale fault features are extracted by the multi-head temporal attention algorithm. The multi-scale fault features are output as integrated features through the adaptive feature fusion layer, and the integrated features are output as fault probability distribution through the fully connected classification layer.

[0056] Step A1: Calculate the average temperature gradient value of the normalized temperature value within a sliding window including the current time point.

[0057] According to the formula Determine the temperature gradient.

[0058] in, for Normalized temperature value at each moment, for Normalized temperature value at each moment, is the current time step, is the sliding index variable, for The average temperature gradient in the window, for Normalized temperature value at each moment.

[0059] Step A2: multiply the absolute value of the average temperature gradient value by a preset gating enhancement coefficient to generate a gradient gating term.

[0060] Step A3: Add the gradient gate term to the original input term of the forget gate to determine the forget gate input.

[0061] Step A4: Apply a sigmoid activation function to the enhanced forget gate input to determine the forget gate output; wherein the forget gate output determines the degree to which the cell state at the previous moment is retained.

[0062] Steps A2-A4 are expressed by the following formula:

[0063] in, , is the normalized temperature value, is the concatenated vector of weighted second-order gradient entropy, β is the gate enhancement coefficient, For the forget gate output, is the output of the forget gate, for Normalized temperature value at each moment, is the sigmoid function, for Normalized temperature value at each moment, is the hidden state at the previous moment.

[0064] Furthermore, a multi-head temporal attention algorithm is used to extract multi-scale fault features, including: B1. Set up multiple independent attention heads.

[0065] Set multiple independent attention heads; wherein the multiple independent attention heads include short window independent attention heads, medium window independent attention heads, and long window independent attention heads.

[0066] B2. Set different time window scales for multiple independent attention heads; the time window scale is the length of the historical time step range that the attention mechanism focuses on.

[0067] B3. Generate a query vector based on the hidden state output of the long short-term memory network layer at the current moment.

[0068] B4. Generate a key vector matrix and a value vector matrix based on the historical hidden state output sequence within the time scale window.

[0069] B5. Calculate the dot product of the query vector and the key vector matrix, divide by the preset scaling factor, perform masking, and apply the softmax function to obtain the weight distribution.

[0070] B6. Perform weighted summation on the value vector matrix based on the weight distribution to determine the output vector of the independent attention head at the current time window scale.

[0071] B2 to B6 are represented by the following formula:

[0072] in, For the The output of the attention head, the Attention function is the standard dot product attention function, and , is the query matrix, For the The query weight matrix of the attention head, is the bond matrix, For the The key weight matrix of the attention head, is the value matrix, For the The value weight matrix of the attention head, is the softmax function, is the scaling factor, is the transpose of the key matrix, is the lower triangular mask matrix, is element-wise multiplication.

[0073] B7. Concatenate the output vectors of multiple independent attention heads to output multi-scale fault features.

[0074] The attention features output by multiple independent attention heads are calculated according to the formula splicing; among them, is the output of multi-head temporal attention, is the feature vector concatenation operation, is the number of attention heads.

[0075] Furthermore, the multi-scale fault features are integrated through the adaptive feature fusion layer to output the following features: C1. Concatenate the current hidden state output of the long short-term memory network with the feature vector output by the multi-head temporal attention algorithm to obtain a concatenated feature vector.

[0076]

[0077] in, is the basic weight matrix, is the weight adjustment coefficient, is the temperature variance in the current time window C2. Process the concatenated feature vectors based on the sigmoid activation function to generate a fused gate value vector.

[0078] Steps C1 and C2 are represented by the following formulas:

[0079] in, is the fusion gate value vector, is the original hidden state of LSTM, is the bias vector of the fusion gate, is the sigmoid function.

[0080] C3. Based on the hidden state output of the long short-term memory network layer and the fusion gate value vector, determine the weighted hidden state features.

[0081] C4, based on the feature vector output by the multi-head temporal attention algorithm and the fused gate value vector to determine the weighted attention feature.

[0082] Step C3 and step C4 are expressed by the following formulas:

[0083] in, is the fused feature vector, is the weighted hidden state feature, is the weighted attention feature.

[0084] C5. Generate integrated features based on the weighted latent state features and weighted attention features.

[0085] The fused features are input into the fully connected layer and output after activation by the softmax function ;in, is the failure probability distribution.

[0086] In a specific case, a device A needs to build an end-to-end fault diagnosis model. First, improve the LSTM layer: calculate the sliding window ( = 20 seconds) (the average temperature gradient (the average of the temperature change rate between the current moment and the previous 20 seconds)) is injected into the forget gate. When the equipment experiences slow bearing wear, this mechanism enables the forget gate to retain a long-term memory of the rising temperature trend, reducing the missed detection rate by 40%. Secondly, a multi-head attention system is implemented: a short window head (10 steps) captures transient impacts in the cooling system, a medium window head (30 steps) identifies abnormal load fluctuations, and a long window head (60 steps) tracks gear wear trends. Each head calculates feature weights using standard dot-product attention, extracting multi-scale information in parallel. Finally, adaptive feature fusion is implemented: based on the temperature variance in the current time window (for example, when the variance is > 2°C), the gated network dynamically increases the weight of the attention feature (for example, from 0.3 to 0.7), addressing the limited adaptability of fixed fusion to sudden failures. The fused features are fed into a fully connected layer to output the fault probability (for example, an 85% probability for "bearing wear").

[0087] like Figure 6 As shown, a model architecture ablation experiment is conducted to verify the necessity of key components in the end-to-end diagnosis model. Figure 6 Dimensional metrics (precision, recall, F1 score, area under the curve, early detection rate, and false alarm rate) were compared between the full model and a simplified version that removed components such as multi-head attention and adaptive fusion. The experimental results show that the full model achieves the largest envelope across all metrics, with particularly significant advantages in early detection rate and false alarm rate. Removing multi-head attention leads to a decrease in multi-scale feature extraction capabilities, with the most significant reduction in the area under the curve. Removing the adaptive fusion layer significantly reduces recall (increases missed detections), demonstrating the critical role of dynamic feature fusion in adaptability to operating conditions. The traditional long short-term memory network baseline model performs the weakest across all dimensions, validating the effectiveness of gradient gating for slowly varying fault memory.

[0088] Step 5: Update the fault prototype vector weight according to the change of the equipment working condition of the equipment to be detected.

[0089] Step 5.1, based on Determine the constrained inter-class distance to determine the prototype contrast loss; where, For the class prototype; is the prototype contrast loss, is the number of fault categories, For the The mean of the class sample features, For category The characteristic output of the sample, is the boundary margin, For category The characteristic output of the sample, For category The characteristic output of the sample, is the Euclidean distance, For category The number of samples, is the sample set of category i, is the sample index, is the feature extraction network, is the nth sample, Represents the maximum value function.

[0090] Step 5.2: Introduce the preset cross entropy loss function.

[0091] Introducing the preset cross entropy loss function .

[0092] Step 5.3: Determine the total loss function based on the prototype contrast loss and the cross entropy loss function.

[0093]

[0094] in, is the total loss function, is the cross entropy loss function, is the balance factor.

[0095] Step 5.4: Determine the new batch data and historical batch data of the equipment to be tested.

[0096] Step 5.5: When the batch difference between the new batch feature distribution of the new batch data and the historical feature distribution of the historical batch data is greater than a preset difference threshold, the update rate is determined according to the value of the batch difference.

[0097]

[0098]

[0099] in, After the update class prototype vector, is the adaptive update rate, is the adaptive update rate, Category in the current batch A sample of For batch Medium sample The input features of is the sample index, is the feature covariance matrix, is the Frobenius norm, is the basic learning rate, is the drift sensitivity coefficient, is an exponential function with natural numbers as base, For historical data sets All samples.

[0100] Step 5.6: Update the fault prototype vector weight based on the update rate.

[0101] In a specific case, the temperature characteristic drift of a device A due to the load configuration upgrade. First, calculate the prototype comparison loss: extract the feature mean of the historical data set (including three types of samples: normal, bearing wear, and cooling failure) as the class prototype (such as the bearing wear prototype ), constraining the characteristics of similar samples in training The distance is less than the heterogeneous distance (margin m=0.5). The total loss function is combined with the cross entropy loss (weight λ=0.8) to improve the clarity of the classification boundary. When the difference between the feature distribution of the new batch data and the historical distribution exceeds the threshold, the incremental update is triggered: the adaptive update rate η is calculated based on the difference (the basic learning rate =0.1, drift sensitivity coefficient =0.3), updated prototype vector Fuse new sample features (e.g. weight η = 0.15).

[0102] Step 6: Determine the confidence score based on the feature deviation, classification uncertainty, and gradient entropy change trend in the end-to-end fault diagnosis model.

[0103] Step 6.1: Measure the Euclidean distance between the current fusion feature vector and the feature prototype vector of the end-to-end fault diagnosis model and divide it by the maximum norm of the feature prototype vectors of all categories to determine the feature deviation.

[0104] The characteristic deviation is given by express; in, is the fused feature vector, For the predicted category The prototype, is the maximum value of all category prototype norms, is the characteristic mean of the samples of category i.

[0105] Step 6.2: Calculate the fault probability distribution output by the end-to-end fault diagnosis model to determine the information entropy and thus the classification uncertainty; wherein, a higher entropy value of the classification uncertainty indicates a greater uncertainty.

[0106]

[0107] in, is the classification uncertainty, Output probability for the model.

[0108] Step 6.3: Determine the gradient entropy change based on the absolute value of the difference between the weighted second-order gradient entropy eigenvalue at the current moment and the second-order gradient entropy eigenvalue at the previous moment.

[0109]

[0110] in, for Time-weighted second-order gradient entropy, for Moment-weighted second-order gradient entropy.

[0111] Step 6.4: Multiply the feature deviation, classification uncertainty, and gradient entropy change by the preset weight coefficients and add them together to determine the weighted comprehensive value.

[0112] Step 6.5: Input the weighted comprehensive value into the sigmoid function for nonlinear mapping to determine the confidence score.

[0113] Steps 6.4 to 6.5 are represented by the following formula:

[0114] in, is the fault confidence score, is the fault confidence score, is the classification uncertainty weight, is the gradient entropy change weight.

[0115] In a specific example, when a device A is diagnosed online, the confidence level is calculated in real time. The feature deviation (the distance ratio between the current feature and the prototype of the "seal failure" class = 0.9), classification uncertainty (the entropy value of the output probability distribution = 0.2), and gradient entropy change ( =0.3). According to the preset weights (w1=0.6, w2=0.3, w3=0.1), the confidence score is generated: sig(0.6×0.9+0.3×0.2+0.1×0.3)=0.85.

[0116] Step 7: Generate feature vectors from the real-time collected temperature data in sequence and input them into the trained end-to-end fault diagnosis model to obtain the fault probability distribution.

[0117] In a specific example, during real-time monitoring of equipment A, the central processing unit performs normalization (window length τ = 30 seconds) and weighted second-order gradient entropy extraction (binning set B = 5 intervals) on the temperature data collected every second. These two are concatenated into a feature vector (dimension 64), which is then input into the end-to-end model trained in step 4. Through LSTM time series modeling and multi-head attention fusion, the output is a multi-fault probability distribution (e.g., "bearing wear" with a 75% probability, "cooling failure" with a 10% probability, and "normal" with a 15% probability).

[0118] Step 8: When the confidence score exceeds the confidence threshold and the failure probability in the failure probability distribution continues to be higher than the preset probability threshold, an early warning is issued.

[0119] In a specific example, the confidence score for the diagnosis of device A was 0.78 (threshold 0.7), and the probability of "compressor failure" was above 85% for 10 consecutive minutes. This triggered a Level 2 alert: an alarm work order was sent to the maintenance terminal, and a backup cooling unit was activated.

[0120] like Figure 7 As shown, a comparison of dynamic adaptability during long-term operation was conducted to verify the dynamic weight update mechanism's adaptability to device state drift. Over a 100-day operating cycle, the accuracy stability of the present invention was compared with that of a fixed model, a conventional long-short-term memory network, and a traditional method under four device state changes (load configuration change, cooling system upgrade, ambient temperature change, and device aging). The fixed model experienced a step-by-step performance degradation after each state change, with the degradation exacerbated during the device aging phase. Conventional long-short-term memory networks were sensitive to ambient temperature changes and exhibited the largest accuracy fluctuations. The present invention's method quickly recovered high accuracy after all four state changes, demonstrating that the dynamic prototype adjustment mechanism, through sensing differences in feature distributions, enables incremental optimization of model parameters, enabling the diagnostic system to continuously evolve.

[0121] The above is an embodiment of the method proposed in this application. Based on the same inventive concept, this application embodiment also provides an equipment warning device based on intelligent diagnosis, whose structure is as follows Figure 2 shown.

[0122] Figure 8 This is a schematic diagram of the internal structure of an intelligent diagnosis-based warning device provided in an embodiment of the present application. Figure 8 As shown, the equipment includes: at least one processor 801; and, a memory 802 communicatively coupled to the at least one processor; The memory 802 stores instructions that can be executed by at least one processor, and the instructions are executed by the at least one processor 801 to enable the at least one processor 801 to: Deploy a high-precision temperature sensor network to the key components of the equipment to be tested to obtain the original temperature time series data of the key components, and send it to the preset central processing unit through the preset industrial Internet of Things gateway; normalize the original temperature time series data through the central processing unit to generate a normalized temperature value; determine the weighted second-order gradient entropy feature based on the normalized temperature value; build an end-to-end fault diagnosis model; among which, the input of the end-to-end fault diagnosis model is the normalized temperature value and the weighted second-order gradient entropy feature, and the long-term dependency is captured by the improved long short-term memory network layer of the end-to-end fault diagnosis model, and the multi-head temporal attention algorithm is used to obtain the weighted second-order gradient entropy feature. Multi-scale fault features are extracted, and the multi-scale fault features are integrated through the adaptive feature fusion layer to output the integrated features. The integrated features are then output as the fault probability distribution through the fully connected classification layer. The weight of the fault prototype vector is updated according to the changes in the equipment operating conditions of the equipment to be tested. The confidence score is determined based on the feature deviation, classification uncertainty and gradient entropy change trend in the end-to-end fault diagnosis model. The real-time collected temperature data is sequentially generated into feature vectors, which are input into the trained end-to-end fault diagnosis model to obtain the fault probability distribution. When the confidence score exceeds the confidence threshold and the fault probability in the fault probability distribution continues to be higher than the preset probability threshold, an early warning is issued.

[0123] Some embodiments of the present application provide corresponding Figure 1 A non-volatile determination machine storage medium for device early warning based on intelligent diagnosis stores determination machine executable instructions, wherein the determination machine executable instructions are set to: Deploy a high-precision temperature sensor network to the key components of the equipment to be tested to obtain the original temperature time series data of the key components, and send it to the preset central processing unit through the preset industrial Internet of Things gateway; normalize the original temperature time series data through the central processing unit to generate a normalized temperature value; determine the weighted second-order gradient entropy feature based on the normalized temperature value; build an end-to-end fault diagnosis model; among which, the input of the end-to-end fault diagnosis model is the normalized temperature value and the weighted second-order gradient entropy feature, and the long-term dependency is captured by the improved long short-term memory network layer of the end-to-end fault diagnosis model, and the multi-head temporal attention algorithm is used to obtain the weighted second-order gradient entropy feature. Multi-scale fault features are extracted, and the multi-scale fault features are integrated through the adaptive feature fusion layer to output the integrated features. The integrated features are then output as the fault probability distribution through the fully connected classification layer. The weight of the fault prototype vector is updated according to the changes in the equipment operating conditions of the equipment to be tested. The confidence score is determined based on the feature deviation, classification uncertainty and gradient entropy change trend in the end-to-end fault diagnosis model. The real-time collected temperature data is sequentially generated into feature vectors, which are input into the trained end-to-end fault diagnosis model to obtain the fault probability distribution. When the confidence score exceeds the confidence threshold and the fault probability in the fault probability distribution continues to be higher than the preset probability threshold, an early warning is issued.

[0124] The various embodiments in this application are described in a progressive manner. Similar portions between the various embodiments can be referenced to each other. Each embodiment focuses on the differences from the other embodiments. In particular, the IoT device and media embodiments are generally similar to the method embodiments, so their description is relatively simple. For relevant portions, refer to the description of the method embodiments.

[0125] The system and medium provided in the embodiments of the present application correspond one-to-one to the method. Therefore, the system and medium also have similar beneficial technical effects to their corresponding methods. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the system and medium will not be repeated here.

[0126] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of a fully hardware embodiment, a fully software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0127] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and determination machine program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by the determination machine program instructions. These determination machine program instructions can be provided to a processor of a general determination machine, a special determination machine, an embedded processing machine, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the determination machine or other programmable data processing device generate instructions for implementing the process Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0128] These determination machine program instructions may also be stored in a determination machine readable memory that can direct a determination machine or other programmable data processing device to operate in a specific manner, so that the instructions stored in the determination machine readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0129] These determination machine program instructions may also be loaded onto a determination machine or other programmable data processing device so that a series of operation steps are executed on the determination machine or other programmable device to produce the processing implemented by the determination machine, thereby providing instructions executed on the determination machine or other programmable device for implementing the process in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0130] In a typical configuration, the determination device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0131] The memory may include non-permanent storage in a certain machine-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a certain machine-readable medium.

[0132] Machine-readable media include permanent and non-permanent, removable and non-removable media that can be used to store information using any method or technology. The information can be machine-readable instructions, data structures, program modules or other data. Examples of machine-readable storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a device. As defined herein, machine-readable media does not include transitory media such as modulated data signals and carrier waves.

[0133] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0134] The above are merely embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A device early warning method based on intelligent diagnosis, characterized in that: The method comprises: Deploy a high-precision temperature sensor network to the key components of the equipment to be tested to obtain the original temperature time series data of the key components and send it to the preset central processing unit through the preset industrial Internet of Things gateway; Normalizing the original temperature time series data by the central processing unit to generate a normalized temperature value; determining a weighted second-order gradient entropy feature based on the normalized temperature value; Constructing an end-to-end fault diagnosis model; wherein the input of the end-to-end fault diagnosis model is a normalized temperature value and a weighted second-order gradient entropy feature, long-term dependencies are captured through an improved long short-term memory network layer of the end-to-end fault diagnosis model, multi-scale fault features are extracted through a multi-head temporal attention algorithm, the multi-scale fault features are output as integrated features through an adaptive feature fusion layer, and the integrated features are output as a fault probability distribution through a fully connected classification layer; Updating the fault prototype vector weight according to the change of the equipment working condition of the equipment to be detected; determining a confidence score based on feature deviation, classification uncertainty, and gradient entropy change trend in the end-to-end fault diagnosis model; The real-time collected temperature data is sequentially generated into feature vectors, which are then input into the trained end-to-end fault diagnosis model to obtain the fault probability distribution. When the confidence score exceeds the confidence threshold and the failure probability in the failure probability distribution continues to be higher than the preset probability threshold, an early warning is issued.

2. The device early warning method based on intelligent diagnosis according to claim 1 is characterized in that: Normalizing the original temperature time series data by the central processing unit to generate a normalized temperature value specifically includes: Setting a time window length parameter, and dividing the continuous overlapping time windows based on the time window length parameter; wherein the time window includes a plurality of data points, and the data points are temperature data corresponding to the time series in the original temperature time series; Determining a sliding mean of the time window based on the arithmetic mean of all data points in the time window; The sliding standard deviation of the data points in the time window is determined by the sum of the squares of the deviations between the data points in the window and the sliding mean; The temperature data of each time series within the time window is calculated based on the sliding mean and the sliding standard deviation to generate a normalized temperature value.

3. The device early warning method based on intelligent diagnosis according to claim 2, characterized in that: Determining a weighted second-order gradient entropy feature based on the normalized temperature value specifically includes: Performing a differential operation based on the temperature values ​​of any three adjacent time points in the normalized temperature value to determine a second-order difference; wherein the second-order difference is used to suppress high-order noise; The calculation of the second-order difference is expressed by the following formula: in, is the second-order difference, for The normalized temperature value at the moment, for The normalized temperature value at the moment, for Normalized temperature value at the moment; Setting a fixed-length window, and dividing the second-order difference into a plurality of gradient value bins according to a preset range within the fixed-length time window to determine a probability distribution of each gradient value bin; Determining a bin weight coefficient based on the mean and maximum gradient of the gradient value bins; The weighted second-order gradient entropy feature is determined based on the probability distribution and the box weight coefficient.

4. The device early warning method based on intelligent diagnosis according to claim 1, characterized in that: The construction of the improved long short-term memory network layer specifically includes: Calculating an average temperature gradient value of the normalized temperature value within a sliding window including the current time point; Multiplying the absolute value of the average temperature gradient value by a preset gating enhancement coefficient to generate a gradient gating term; Adding the gradient gating term to the original input term of the forget gate to determine the forget gate input; A sigmoid activation function is applied to the enhanced forget gate input to determine the forget gate output; wherein the forget gate output determines the degree to which the cell state at the previous moment is retained.

5. The device early warning method based on intelligent diagnosis according to claim 4 is characterized in that: The multi-head temporal attention algorithm is used to extract multi-scale fault features, including: Set up multiple independent attention heads; Setting different time window scales for multiple independent attention heads; where the time window scale is the length of the historical time step range that the attention mechanism focuses on; Generate a query vector based on the hidden state output of the long short-term memory network layer at the current moment; Generate a key vector matrix and a value vector matrix based on the historical hidden state output sequence within the time scale window; Calculate the dot product of the query vector and the key vector matrix, divide by a preset scaling factor, perform masking, and apply a softmax function to obtain a weight distribution; Performing a weighted summation on the value vector matrix based on the weight distribution to determine an output vector of the independent attention head at the current time window scale; The output vectors of multiple independent attention heads are concatenated to output multi-scale fault features.

6. The device early warning method based on intelligent diagnosis according to claim 5 is characterized in that: The multi-scale fault features are integrated by the adaptive feature fusion layer to output the integrated features, specifically including: Concatenate the current hidden state output of the long short-term memory network with the feature vector output by the multi-head temporal attention algorithm to obtain a concatenated feature vector; Processing the concatenated feature vector based on a sigmoid activation function to generate a fused gate value vector; Determining weighted hidden state features based on the hidden state output of the long short-term memory network layer and the fused gate value vector; Determine a weighted attention feature based on the feature vector output by the multi-head temporal attention algorithm and the fused gate value vector; An integrated feature is generated based on the weighted latent state feature and the weighted attention feature.

7. The device early warning method based on intelligent diagnosis according to claim 6, characterized in that: Updating the fault prototype vector weight according to the change in the operating condition of the device to be detected specifically includes: based on Determine the constrained inter-class distance to determine the prototype contrast loss; where, For the class prototype; is the prototype contrast loss, is the number of fault categories, For the The mean of the class sample features, For category The characteristic output of the sample, is the boundary margin, For category The characteristic output of the sample, For category The characteristic output of the sample, is the Euclidean distance, For category The number of samples, is the sample set of category i, is the sample index, is the feature extraction network, is the nth sample, represents the maximum value function; Introducing a preset cross entropy loss function; The total loss function is determined based on the prototype contrast loss and the cross entropy loss function: Determining new batch data and historical batch data of the device to be tested; When the batch difference between the new batch feature distribution of the new batch data and the historical feature distribution of the historical batch data is greater than a preset difference threshold, determining an update rate according to the value of the batch difference; The fault prototype vector weight is updated based on the update rate.

8. The device early warning method based on intelligent diagnosis according to claim 1 is characterized in that: Determining a confidence score based on feature deviation, classification uncertainty, and gradient entropy change trend in the end-to-end fault diagnosis model, specifically including: Measuring the Euclidean distance between the current fusion feature vector and the feature prototype vector of the end-to-end fault diagnosis model and dividing the distance by the maximum norm of the feature prototype vectors of all categories to determine the feature deviation; Calculating the fault probability distribution output by the end-to-end fault diagnosis model to determine information entropy to determine classification uncertainty; wherein a higher entropy value of the classification uncertainty indicates greater uncertainty; Determine the gradient entropy change based on the absolute value of the difference between the weighted second-order gradient entropy eigenvalue at the current moment and the second-order gradient entropy eigenvalue at the previous moment; The feature deviation, classification uncertainty and gradient entropy change are multiplied by preset weight coefficients respectively and then added to determine a weighted comprehensive value; The weighted comprehensive value is input into a sigmoid function for nonlinear mapping to determine a confidence score.

9. An equipment early warning device based on intelligent diagnosis, characterized in that: The device comprises: at least one processor; and, a memory communicatively coupled to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: Deploy a high-precision temperature sensor network to the key components of the equipment to be tested to obtain the original temperature time series data of the key components and send it to the preset central processing unit through the preset industrial Internet of Things gateway; Normalizing the original temperature time series data by the central processing unit to generate a normalized temperature value; determining a weighted second-order gradient entropy feature based on the normalized temperature value; Constructing an end-to-end fault diagnosis model; wherein the input of the end-to-end fault diagnosis model is a normalized temperature value and a weighted second-order gradient entropy feature, long-term dependencies are captured through an improved long short-term memory network layer of the end-to-end fault diagnosis model, multi-scale fault features are extracted through a multi-head temporal attention algorithm, the multi-scale fault features are output as integrated features through an adaptive feature fusion layer, and the integrated features are output as a fault probability distribution through a fully connected classification layer; Updating the fault prototype vector weight according to the change of the equipment working condition of the equipment to be detected; determining a confidence score based on feature deviation, classification uncertainty, and gradient entropy change trend in the end-to-end fault diagnosis model; The real-time collected temperature data is sequentially generated into feature vectors, which are then input into the trained end-to-end fault diagnosis model to obtain the fault probability distribution. When the confidence score exceeds the confidence threshold and the failure probability in the failure probability distribution continues to be higher than the preset probability threshold, an early warning is issued.

10. A non-volatile determination machine storage medium for device early warning based on intelligent diagnosis, storing determination machine executable instructions, characterized in that: The determination machine executable instruction is set to: Deploy a high-precision temperature sensor network to the key components of the equipment to be tested to obtain the original temperature time series data of the key components and send it to the preset central processing unit through the preset industrial Internet of Things gateway; Normalizing the original temperature time series data by the central processing unit to generate a normalized temperature value; determining a weighted second-order gradient entropy feature based on the normalized temperature value; Constructing an end-to-end fault diagnosis model; wherein the input of the end-to-end fault diagnosis model is a normalized temperature value and a weighted second-order gradient entropy feature, long-term dependencies are captured through an improved long short-term memory network layer of the end-to-end fault diagnosis model, multi-scale fault features are extracted through a multi-head temporal attention algorithm, the multi-scale fault features are output as integrated features through an adaptive feature fusion layer, and the integrated features are output as a fault probability distribution through a fully connected classification layer; Updating the fault prototype vector weight according to the change of the equipment working condition of the equipment to be detected; determining a confidence score based on feature deviation, classification uncertainty, and gradient entropy change trend in the end-to-end fault diagnosis model; The real-time collected temperature data is sequentially generated into feature vectors, which are then input into the trained end-to-end fault diagnosis model to obtain the fault probability distribution. When the confidence score exceeds the confidence threshold and the failure probability in the failure probability distribution continues to be higher than the preset probability threshold, an early warning is issued.

Images