Edge diagnostic model algorithm driven intelligent diagnosis method for major equipment failure

Abstract

The application relates to the technical field of equipment fault diagnosis, and discloses a major equipment fault intelligent diagnosis method driven by an edge diagnosis model algorithm. Real-time monitoring data streams in the operation process of major equipment are collected, wherein the real-time monitoring data streams contain vibration signals, temperature readings and sound wave characteristics, and device operation state waveform atlas and device state diagnosis auxiliary documents are generated according to the data streams; the generated waveform atlas and auxiliary documents are input into a pre-configured edge diagnosis model for processing, and initial classification data of corresponding fault types of the device are obtained; a plurality of fault mechanism characteristic parameters are extracted based on the waveform atlas, matched with a pre-defined fault mechanism model set, and auxiliary classification data is generated; the initial classification data and the auxiliary classification data are fused to determine a final fault type, and the fault root cause is inferred in combination with the final fault type. Through multi-dimensional data fusion and model combination, the method realizes accurate diagnosis and root cause tracing of major equipment faults.

Description

Technical Field

[0001] This invention relates to the field of equipment fault diagnosis technology, specifically to an intelligent diagnostic method for major equipment faults driven by an edge diagnostic model algorithm. Background Technology

[0002] In critical sectors such as industrial production, energy supply, and transportation, the stable operation of major equipment is essential for production safety and efficiency. This equipment often has a complex structure and numerous interconnected components. Its failures are characterized by their high degree of concealment, complex causes, and wide-ranging impact. Once a failure occurs, it may lead to production interruption, economic losses, or even safety accidents.

[0003] Traditional equipment fault diagnosis methods largely rely on manual inspections and periodic maintenance. Manual inspections require professionals to judge equipment status based on experience, which not only consumes a lot of manpower but is also limited by individual experience and observation angles, making it difficult to capture early, subtle fault characteristics and easily leading to missed potential faults. Periodic maintenance, on the other hand, has the limitation of fixed cycles, and failures may occur unexpectedly during maintenance intervals, making it impossible to achieve dynamic monitoring of equipment status.

[0004] With the development of sensing and computer technologies, some devices have begun to use sensors to collect operational data and transmit it to a centralized platform for analysis and diagnosis. However, in this centralized architecture, the transmission of massive amounts of real-time monitoring data consumes a significant amount of network bandwidth, especially in scenarios where devices are widely distributed, where network latency becomes a prominent issue, making it difficult to meet the real-time requirements of fault diagnosis. At the same time, the centralized platform faces considerable computational pressure and is prone to processing delays during peak data periods, impacting diagnostic efficiency.

[0005] The rise of edge computing technology has provided new ideas for equipment diagnostics. By deploying edge nodes at the device end to achieve localized data processing, it reduces data transmission volume and latency. However, existing edge computing-based diagnostic models mostly rely on a single type of monitoring data (such as vibration signals or temperature data) for fault identification, ignoring the correlation between different types of data. For example, some faults may simultaneously manifest as abnormal vibration and elevated temperature; relying on only a single data point leads to an incomplete description of the fault characteristics, affecting diagnostic accuracy.

[0006] Existing diagnostic models primarily focus on fault type classification, lacking integration with equipment failure mechanisms. Fault mechanisms reflect the failure patterns of equipment components under specific operating conditions and are the intrinsic causes of faults. Diagnostic results divorced from fault mechanisms often only identify fault phenomena, failing to trace the root cause. When different faults exhibit similar monitoring characteristics, misjudgments are prone to occur, and clear guidance for fault handling cannot be provided. Furthermore, the separation of fault type identification from root cause inference increases the complexity of fault handling and prolongs equipment recovery time. Summary of the Invention

[0007] The purpose of this invention is to provide an intelligent diagnostic method for major equipment faults driven by an edge diagnostic model algorithm, so as to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, this invention provides an intelligent diagnostic method for critical equipment faults driven by an edge diagnostic model algorithm, the method comprising: Collect real-time monitoring data streams during the operation of critical equipment. The real-time monitoring data streams include vibration signals, temperature readings, and acoustic characteristics. Generate equipment operating status waveform diagrams and equipment status diagnosis auxiliary documents based on the real-time monitoring data streams. The waveform diagram of the equipment's operating status and the auxiliary document for equipment status diagnosis are input into a pre-configured edge diagnosis model for processing to obtain initial classification data of the fault types corresponding to the critical equipment. Based on the waveform spectrum of the equipment's operating status, multiple fault mechanism feature parameters of the critical equipment are extracted and matched with a predefined set of fault mechanism models. Based on the result of the matching operation, auxiliary classification data corresponding to the fault type of the critical equipment is generated. By integrating the initial classification data and the auxiliary classification data, the final failure type of the critical equipment is determined, and the root cause of the failure of the critical equipment is inferred based on the final failure type.

[0009] Preferably, the step of generating the device operating status waveform diagram and device status diagnostic auxiliary document based on the real-time monitoring data stream includes: Based on real-time monitoring data stream, a waveform spectrum of equipment operation status is constructed, and multiple fault mechanism feature components are extracted from the real-time monitoring data stream. The fault mechanism feature components cover amplitude dispersion, phase concentration, polarity effect performance, flight trajectory characteristics, pulse balance, low-frequency component characteristics, and high-frequency component characteristics. The polarity effect is determined by analyzing the difference in energy distribution of the signal in the positive and negative half-cycles; the flight trajectory characteristics are obtained by plotting trajectory curves from continuously acquired displacement signals, and then analyzing the curvature changes and offset trends of the curves. Quantify the correlation between real-time monitoring data streams and environmental monitoring data to generate external impact factors of faults; Each fault mechanism feature component and external influencing factor is discretized and converted into text, and the processing results are represented in key-value pair form to form an auxiliary document for equipment status diagnosis.

[0010] Preferably, the environmental monitoring data includes real-time temperature data and real-time humidity data; The correlation between the quantified real-time monitoring data stream and the environmental monitoring data is used to generate external impact factors of the fault, including: Dynamic covariance calculation is performed on real-time monitoring data streams and real-time temperature data within the same time period to obtain the temperature correlation coefficient; Dynamic covariance calculation is performed on real-time monitoring data streams and real-time humidity data within the same time period to obtain the humidity correlation coefficient; The temperature correlation coefficient and humidity correlation coefficient are weighted and superimposed according to a preset weight ratio to output the external influence factor of the fault.

[0011] Preferably, the pre-configured edge diagnostic model includes a temporal coding module, a text parsing module, a multimodal fusion module, and a multimodal output module; The process involves inputting the equipment operating status waveform graph and the equipment status diagnostic auxiliary document into a pre-configured edge diagnostic model for processing to obtain initial classification data for the corresponding fault types of the critical equipment, including: The timing coding module encodes the waveform of the device's operating status to generate waveform feature vectors. The text parsing module encodes the device status diagnostic auxiliary document to generate a text feature vector. Based on the cross-modal alignment loss function and the multimodal matching loss function, the waveform feature vector and the text feature vector are input into the multimodal fusion module for processing to obtain the multimodal fusion result; The multimodal fusion results are input into the multimodal output module for processing, and the initial classification data corresponding to the fault types of critical equipment is output.

[0012] Preferably, the step of extracting multiple fault mechanism feature parameters of the critical equipment based on the waveform spectrum of the equipment's operating status, and performing a matching operation with a predefined set of fault mechanism models, and generating auxiliary classification data corresponding to the fault type of the critical equipment based on the result of the matching operation, includes: The waveform spectrum of the equipment's operating status is quantified by multiple fault mechanism characteristic parameters, including amplitude level, abnormal event count, event interval duration, comparison of low-frequency and high-frequency components, positive and negative half-axis symmetry characteristics, single-peak or double-peak performance, fractal characteristics, ultrasonic detection probability, and pulse sequence characteristics. Multiple fault mechanism feature parameters are matched with a set of predefined fault mechanism models corresponding to various fault types to generate multiple matching degree values. The fault type associated with the fault mechanism model corresponding to the highest matching degree value is used as auxiliary classification data.

[0013] Preferably, the initial classification data includes fault types and their corresponding probability values; The process of integrating the initial classification data and the auxiliary classification data to determine the final failure type of the critical equipment includes: The weight factors for the initial classification data are calculated based on the probability values, and the auxiliary weight factors for the auxiliary classification data are calculated based on the matching degree values. The weighting factors and auxiliary weighting factors are normalized to generate normalized weighting factors; By combining normalized weighting factors, initial classification data, and auxiliary classification data, the final failure type of critical equipment is determined.

[0014] Preferably, the inference of the root cause of failure of critical equipment includes: Construct a cross-validation rule base, which contains the mapping relationship between fault modes recorded in the historical fault database and environmental monitoring data; The final fault type is combined with the current environmental monitoring data to form a verification feature pair; Traverse all validation rules in the cross-validation rule base and calculate the matching confidence of the validation feature pair with each validation rule; Extract the fault root cause pattern associated with the verification rule corresponding to the highest matching confidence, and use it as the fault root cause determination result of the critical equipment.

[0015] Preferably, after inferring the root cause of the failure of the critical equipment, the method further includes: Construct a set of potential defect features for critical equipment, including excessively high ambient humidity, signs of component aging, and familial genetic defects; Obtain equipment attribute information of major equipment, and combine the equipment attribute information with environmental monitoring data to determine whether the major equipment meets the potential defect feature set, and generate the probability of occurrence of each potential defect feature to form a risk identification list; Early warning signals for critical equipment are generated based on the risk identification list.

[0016] Preferably, generating the probability of occurrence of each potential defect feature to form a risk identification list includes: Decompose the device attribute information and extract multiple attribute feature sub-values; The attribute feature sub-values ​​are input into a pre-trained risk assessment model to generate the probability of occurrence of each potential defect feature; Summarize the probability of occurrence of each potential defect feature to construct a risk identification list.

[0017] Preferably, the step of inputting attribute feature sub-values ​​into a pre-trained risk assessment model to generate the occurrence probability of each potential defect feature includes: Normalize the attribute feature sub-values ​​to generate standardized feature values; Based on standardized feature values, the contribution weight of each potential defect feature is calculated; By combining contribution weights and standardized feature values, the probability of occurrence of each potential defect feature is generated.

[0018] Compared with the prior art, the beneficial effects of the present invention are: This intelligent diagnostic method for critical equipment faults incorporates multiple types of real-time monitoring data, including vibration signals, temperature readings, and acoustic characteristics, during the data acquisition phase. It generates waveform graphs of the equipment's operating status and auxiliary documents for equipment status diagnosis, comprehensively capturing various characteristics of equipment operation. Different types of data reflect the equipment's status from different perspectives, complementing each other and avoiding the limitations of a single data type in describing the equipment's condition. This provides a rich information foundation for subsequent diagnosis.

[0019] By processing waveform maps and auxiliary documents using a pre-configured edge diagnostic model to obtain initial classification data, and leveraging the localized processing characteristics of edge computing, real-time data analysis is achieved. This localized processing reduces the process of transmitting data to remote locations, mitigates the impact of network latency on diagnostic timeliness, and ensures that diagnostic results closely follow changes in equipment operating status, meeting the real-time diagnostic needs of critical equipment. Simultaneously, the edge node processing method reduces dependence on network bandwidth, avoiding diagnostic lag issues caused by data congestion in centralized processing.

[0020] Multiple fault mechanism feature parameters are extracted based on waveform spectra and matched with a predefined set of fault mechanism models to generate auxiliary classification data. This process combines data-driven analysis with the equipment's own fault mechanisms. The fault mechanism feature parameters originate from the equipment's structural characteristics and operating principles. The matching results with the predefined models verify and supplement the initial classification data from the perspective of the equipment's inherent laws, making the fault classification more closely aligned with the equipment's actual operating mechanism and reducing the potential problem of disconnect between the data-driven model and the actual mechanism.

[0021] The fusion of initial and auxiliary classification data combines the advantages of both classification methods, making the final fault type more reliable. This fusion is not a simple addition of results, but rather a process of mutual verification through multi-dimensional information, reducing the risk of misjudgment that might arise from a single model or a single data type. For critical equipment faults with complex correlation characteristics, this fusion process can more accurately identify their specific types, reducing the possibility of missed or incorrect diagnoses.

[0022] The process of inferring the root cause of a fault by combining the final fault type integrates fault diagnosis and root cause tracing into a coherent flow, avoiding the process fragmentation caused by the separation of the two in traditional methods. By directly inferring the correlation between fault type and root cause, the fundamental cause of the fault can be quickly located based on the determined fault type, combined with the structural relationships and operating mechanisms of the equipment. This provides a clear direction for timely fault handling and meets the efficiency and accuracy requirements of handling major equipment faults. Attached Figure Description

[0023] Figure 1 This is a schematic diagram illustrating the working principle of the intelligent diagnostic method for major equipment faults driven by the edge diagnostic model algorithm described in this invention. Figure 2 A flowchart for generating waveform diagrams of equipment operating status and auxiliary documents for equipment status diagnosis; Figure 3 A flowchart for processing device operating status waveform diagrams and device status diagnostic auxiliary documents for edge diagnostic models; Figure 4 A flowchart for determining the final fault type. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Please see Figure 1 This invention provides an intelligent diagnostic method for critical equipment faults driven by an edge diagnostic model algorithm, the method comprising: Collect real-time monitoring data streams during the operation of critical equipment. The real-time monitoring data streams include vibration signals, temperature readings, and acoustic characteristics. Generate equipment operating status waveform diagrams and equipment status diagnosis auxiliary documents based on the real-time monitoring data streams. The equipment operation status waveform diagram and equipment status diagnosis auxiliary document are input into the pre-configured edge diagnosis model for processing to obtain the initial classification data of the fault types corresponding to major equipment. Based on the waveform spectrum of equipment operation status, multiple fault mechanism feature parameters of major equipment are extracted and matched with a predefined set of fault mechanism models. Based on the results of the matching operation, auxiliary classification data corresponding to the fault type of major equipment is generated. By integrating initial classification data and auxiliary classification data, the final failure type of critical equipment is determined, and the root cause of the failure is inferred from the final failure type.

[0026] Example 1: See Figure 2 Based on the real-time monitoring data stream, the system generates equipment operating status waveform diagrams and equipment status diagnosis auxiliary documents. The specific process is as follows: Based on the real-time monitoring data stream, the system constructs equipment operating status waveform diagrams and extracts multiple fault mechanism feature components from the real-time monitoring data stream. The fault mechanism feature components cover amplitude dispersion, phase concentration, polarity effect performance, flight trajectory characteristics, pulse balance, low-frequency component characteristics, and high-frequency component characteristics. Amplitude dispersion is calculated by summing the squares of the deviations of all amplitude values ​​from the average amplitude value within the monitoring period and then dividing by the number of monitoring data points; phase concentration is determined by statistically analyzing the distribution density of phase values ​​around the reference phase at each moment, with a higher value indicating a more concentrated distribution; polarity effect is determined by analyzing the energy distribution differences of the signal in the positive and negative half-cycles, with a more pronounced polarity effect indicating a greater energy difference; flight trajectory characteristics are obtained by plotting trajectory curves from continuously acquired displacement signals and then analyzing the curvature changes and offset trends of the curves; pulse balance is determined by comparing the amplitude differences and interval uniformity of continuous pulse signals; low-frequency component characteristics are obtained by filtering the original signal, extracting signal components with frequencies below a preset threshold, and then calculating the average amplitude and duration of these components; high-frequency component characteristics are obtained by extracting signal components with frequencies above a preset threshold and analyzing the frequency of their peak occurrence and attenuation rate.

[0027] Environmental monitoring data includes real-time temperature and humidity data. The correlation between real-time monitoring data streams and environmental monitoring data is quantified to generate external impact factors for faults. The specific process is as follows: Dynamic covariance calculation is performed on the real-time monitoring data streams and real-time temperature data for the same time period to obtain the temperature correlation coefficient. The dynamic covariance calculation uses a fixed time window as the unit. The length of each time window is set according to the equipment's operating characteristics. Within each window, the mean of the real-time monitoring data stream and the real-time temperature data are calculated separately. Then, the average of the products of the deviations of each data stream from their respective mean values ​​is calculated. The results of all time windows are then averaged to obtain the final temperature correlation coefficient. Similarly, dynamic covariance calculation is performed on the real-time monitoring data streams and real-time humidity data for the same time period to obtain the humidity correlation coefficient. The calculation method is the same as for the temperature correlation coefficient, only the temperature data is replaced with humidity data. The temperature and humidity correlation coefficients are then weighted and superimposed according to a preset weight ratio to output the external impact factors for faults. The preset weight ratio is determined based on the equipment's sensitivity to different environmental factors. If the equipment is more sensitive to temperature changes, the weight of the temperature correlation coefficient is higher; conversely, the weight of the humidity correlation coefficient is higher.

[0028] Each fault mechanism feature component and external influencing factor is discretized and converted into text, with the processing results represented in key-value pair format to form an auxiliary document for equipment condition diagnosis. Discretization employs an equidistant partitioning method, dividing the value range of each feature component into several continuous intervals, each interval corresponding to a discrete label. For example, amplitude dispersion is divided into low, medium, and high intervals, corresponding to labels A, B, and C respectively. External influencing factors are similarly divided into several intervals, each corresponding to a different discrete label. Text conversion combines the name of each feature component and its corresponding discrete label into a text description. For example, when amplitude dispersion is low, the text description is "Amplitude Dispersion: A". The key-value pair format uses the feature component name as the key and the corresponding text description as the value, such as "Amplitude Dispersion": "Amplitude Dispersion: A". The processing results of all fault mechanism feature components and external influencing factors are integrated into a unified format to form a complete auxiliary document for equipment condition diagnosis. Each key-value pair in the document is on a separate line, with a specific symbol separating the key and value, facilitating recognition and processing by the subsequent text parsing module. When generating waveform graphs of equipment operation status, three curves are plotted on the same coordinate system: time on the horizontal axis and the values ​​of vibration signals, temperature readings, and acoustic characteristics from the real-time monitoring data stream on the vertical axis. Different colors and line types are used to distinguish different types of signals. The time span of the waveform graph can be adjusted according to actual monitoring needs, displaying both short-term, real-time updated data and long-term data covering a certain historical period. Scrolling through the graph allows for observation of equipment operation status over different time periods. The graph also highlights points where signals exceed normal ranges with special markers, providing an intuitive visual reference for subsequent fault diagnosis.

[0029] Example 2: See Figure 3The pre-configured edge diagnostic model includes a time-series encoding module, a text parsing module, a multimodal fusion module, and a multimodal output module. The equipment operating status waveform map and equipment status diagnostic auxiliary documents are input into the pre-configured edge diagnostic model for processing to obtain initial classification data for the corresponding fault types of critical equipment. The specific process is as follows: The time-series encoding module encodes the equipment operating status waveform map to generate waveform feature vectors. The time-series encoding module includes an input layer, a hidden layer, and an output layer. The input layer receives the raw data of the waveform map, which is presented in time-series form, containing values ​​of vibration signals, temperature readings, and acoustic characteristics at different times. The hidden layer consists of multiple recurrent units. Each recurrent unit retains the calculation state of the previous time step and combines it with the input data of the current time step to capture the dynamic features of the waveform map that change over time. During processing, the recurrent units transform the waveform data at each time point into a series of intermediate vectors. These intermediate vectors are continuously updated over time and finally integrated by the output layer into a fixed-length waveform feature vector. This vector contains the key feature information of the waveform map throughout the entire time series.

[0030] The text parsing module encodes the equipment status diagnosis auxiliary document to generate a text feature vector. The document exists in key-value pair format, containing discretized results of multiple fault mechanism feature components and external influencing factors. The text parsing module first segments each key-value pair in the document, breaking down the text information into multiple basic semantic units, each corresponding to a word or phrase. Then, an embedding layer transforms these semantic units into low-dimensional vectors, reflecting the semantic relationships between words. The parameters of the embedding layer are obtained through pre-training on a large amount of text data, enabling the mapping of semantically similar words to nearby vector space positions. Following the embedding layer, multiple attention mechanism layers are set up. Each attention mechanism layer assigns weights to the vector representations of different key-value pairs, focusing on information more critical to fault diagnosis. Through multi-layered attention mechanisms, the text parsing module can capture the dependencies between different features in the document, ultimately condensing the entire document's information into a fixed-length text feature vector that accurately reflects the core content of the equipment status diagnosis auxiliary document.

[0031] Based on the cross-modal alignment loss function and the multimodal matching loss function, waveform feature vectors and text feature vectors are input into the multimodal fusion module for processing to obtain the multimodal fusion result. The multimodal fusion module initially integrates the two feature vectors in a concatenated manner to form a joint feature vector. The cross-modal alignment loss function calculates the distance between the waveform feature vector and the text feature vector in the semantic space. When the two vectors express inconsistent semantics, the value of this loss function increases, prompting the model to adjust parameters during training to reduce this inconsistency. The multimodal matching loss function compares the differences between feature vectors under the same device state and feature vectors under different device states. By increasing the distance between different classes and decreasing the distance between the same class, it enhances the model's ability to distinguish different fault types. Under the combined effect of these two loss functions, the multimodal fusion module further transforms and optimizes the joint feature vector. Through fully connected layers and activation functions, the joint feature vector is transformed into a more discriminative multimodal fusion result, which includes both the temporal features of the waveform spectrum and the semantic features of the text document.

[0032] The multimodal fusion results are input to the multimodal output module for processing, outputting initial classification data for the corresponding fault types of critical equipment. The multimodal output module consists of multiple fully connected layers, each containing several neurons connected by weights. The multimodal fusion results first enter the first fully connected layer, where each neuron performs a weighted summation of the input features and a non-linear transformation using an activation function to generate new feature representations. These new feature representations are then passed to the next fully connected layer for similar processing. After multiple transformations, the final feature representations are input to the output layer. The number of neurons in the output layer matches the preset number of fault types, with each neuron corresponding to one fault type. The output layer processes the neuron output values ​​using a softmax function, converting them into probability values. The sum of these probability values ​​is 1, and each probability value represents the likelihood of the critical equipment experiencing the corresponding fault type. The final output initial classification data contains all fault types and their corresponding probability values, providing a quantitative basis for subsequent fault type determination. The entire process is completed on edge computing nodes, enabling rapid response to changes in real-time monitoring data and timely diagnosis of critical equipment faults.

[0033] Example 3: See Figure 4This method extracts multiple fault mechanism feature parameters from the waveform spectrum of equipment operation status, and matches them with a predefined set of fault mechanism models. Based on the matching results, auxiliary classification data corresponding to the fault types of major equipment is generated. The specific process is as follows: Multiple fault mechanism feature parameters from the waveform spectrum of equipment operation status are quantified. These parameters include amplitude level, abnormal event count, event interval duration, comparison of low-frequency and high-frequency components, positive and negative half-axis symmetry characteristics, single-peak or double-peak performance, fractal features, ultrasonic detection probability, and pulse sequence features. Amplitude level is calculated by the difference between the peak and trough values ​​of vibration signals, temperature readings, and acoustic features at each moment in the waveform spectrum, and the maximum difference over multiple periods is taken as the final result. Abnormal event count is calculated by setting a normal fluctuation threshold and counting the number of signal fluctuations exceeding this threshold in the waveform spectrum. Each fluctuation is counted as an abnormal event, starting from when the signal exceeds the threshold and ending when it returns below the threshold. Event interval duration is calculated by recording the start time of two consecutive abnormal events, calculating the difference between them, and averaging all differences as the final result. The comparison of low-frequency and high-frequency components involves spectral analysis of the waveform spectrum, dividing it into low-frequency and high-frequency ranges, calculating the total energy of the signal in each range, and then calculating the ratio of the total energy in the low-frequency range to the total energy in the high-frequency range. The positive and negative half-axis symmetry characteristic is determined by dividing the waveform spectrum along the time axis into positive and negative half-cycles, calculating the area enclosed by the signal and the time axis in each half-cycle, comparing the numerical difference between the two areas, and observing the similarity of the waveform shapes in the two half-cycles. A quantified result is obtained by combining both factors. Single-peak or double-peak representation is determined by scanning the peak values ​​in the waveform spectrum. A peak value significantly higher than the surrounding signal value is identified as a single peak, while two distinct peak values ​​with similar signal values ​​at a certain time interval are identified as double peaks. Fractal characteristics are determined by fractal analysis of the waveform spectrum, calculating the fractal dimension of the waveform. This dimension reflects the complexity of the waveform; a larger value indicates a more complex waveform. The ultrasonic detection probability is calculated using a probability model based on the reflected signal from the equipment surface collected by the ultrasonic sensor, combined with the equipment material characteristics and the detection distance. Pulse sequence characteristics are derived by analyzing the frequency changes, amplitude attenuation patterns, and time intervals between adjacent pulses of the pulse signal to extract quantitative indicators that reflect the pulse characteristics.

[0034] Multiple fault mechanism feature parameters are matched with a predefined set of fault mechanism models corresponding to various fault types, generating multiple matching degree values. Each fault mechanism model for a fault type includes a standard range for each feature parameter, determined based on historical fault data and equipment design parameters. During the matching operation, the deviation of each fault mechanism feature parameter from the corresponding standard range in the model is calculated; the smaller the deviation, the greater the contribution of that parameter to the matching degree. The deviations of all feature parameters are combined to obtain the matching degree value corresponding to that fault type. The fault type associated with the fault mechanism model corresponding to the highest matching degree value is used as auxiliary classification data.

[0035] The initial classification data includes fault types and their corresponding probability values. The initial classification data and auxiliary classification data are integrated to determine the final fault type of critical equipment. The process is as follows: Calculate the weight factor of the initial classification data based on the probability values; the higher the probability value, the larger the weight factor. Calculate the auxiliary weight factor of the auxiliary classification data based on the matching degree value; the higher the matching degree value, the larger the auxiliary weight factor. Normalize the weight factors and auxiliary weight factors to generate normalized weight factors. During normalization, the weight factors and auxiliary weight factors are divided by their sum, using the following formula:

[0036] Where ω represents the normalized weight factor, a represents the weight factor of the initial classification data, and b represents the auxiliary weight factor of the auxiliary classification data.

[0037] By combining normalized weighting factors, initial classification data, and auxiliary classification data, the final failure type of critical equipment is determined. Specifically, for each failure type in the initial classification data, its probability value is multiplied by the normalized weighting factor to obtain a weighted score for that failure type. For each failure type in the auxiliary classification data, its corresponding matching degree value is multiplied by (1 - normalized weighting factor) to obtain a weighted score for that failure type. The weighted scores of all failure types are then summed, and the failure type with the highest score is the final failure type of the critical equipment.

[0038] When calculating amplitude levels, multiple cycles of equipment operation must be covered to ensure that amplitude changes under different operating conditions are captured. During abnormal event counting, threshold settings must refer to the normal operating parameters of the equipment; different models and applications have different thresholds. The calculation of event interval duration must exclude abnormal events during equipment startup and shutdown phases, as signal fluctuations during these phases may not reflect the true fault state of the equipment. The frequency range division for comparing low-frequency and high-frequency components must be determined based on the equipment's operating frequency. For rotating equipment, the low-frequency band is typically set to below 1 / 3 of the operating frequency, and the high-frequency band is set to above 3 times the operating frequency. The similarity of the positive and negative half-axis symmetry characteristics is obtained by calculating the correlation coefficient of the two half-cycle waveforms; the closer the correlation coefficient is to 1, the more similar the shapes. The determination of single-peak or double-peak performance requires excluding noise interference in the signal, and peak identification is performed after smoothing the waveform spectrum. The fractal dimension of the fractal features is calculated using box counting. By covering the waveform with boxes of different sizes, the number of boxes required to cover the waveform is counted, and then the fractal dimension is calculated. The calculation of ultrasonic detection probability needs to consider the sensor's accuracy and detection angle, assigning different weights to detection signals at different locations. Analysis of pulse sequence characteristics requires distinguishing between normal operating pulses and fault pulses. Normal operating pulses typically have stable frequency and amplitude, while fault pulses exhibit irregular variations.

[0039] During the matching process, the fault mechanism model set needs to be updated regularly to incorporate new fault cases and equipment operation data to improve matching accuracy. Each fault type may correspond to multiple fault mechanism models, each applicable to different operating stages and conditions of the equipment. The matching degree calculation requires assigning different weights to different feature parameters, with higher weights given to features that significantly influence the fault type. The probability value of the initial classification data is directly provided by the output layer of the edge diagnostic model; this probability value reflects the model's confidence in the fault type. The matching degree value of the auxiliary classification data ranges from 0 to 1, with values ​​closer to 1 indicating a higher degree of matching. During the fusion process, if the initial classification data and the auxiliary classification data point to the same fault type, the weighted score for that type will significantly increase, making it more likely to be identified as the final fault type. If they point to different fault types, the final result is determined by comparing the weighted scores, avoiding excessive influence of bias from a single classification data point on the diagnostic results.

[0040] Example 4: Inferring the root cause of critical equipment failures. The specific process is as follows: A cross-validation rule base is constructed, containing the mapping relationship between failure modes recorded in the historical failure database and environmental monitoring data. The historical failure database collects past failure cases of the equipment. Each case includes detailed information about the failure, such as the failure type, specific manifestations, occurrence time, duration, and environmental data at the time. By analyzing these cases, the correlation patterns between failure modes and environmental monitoring data are extracted, forming entries in the rule base. Each rule includes the failure type, the environmental data range, and the corresponding root cause. The environmental data range clarifies the specific temperature and humidity ranges when the failure mode occurred, while the root cause describes the underlying cause of the failure, such as component wear, loose connections, or material aging.

[0041] The final fault type is combined with the current environmental monitoring data to form a verification feature pair. The current environmental monitoring data includes real-time temperature and humidity values. These data, together with the final fault type, form a triple containing the fault type, real-time temperature, and real-time humidity, which serves as the verification feature pair for subsequent matching operations.

[0042] The cross-validation rule base is traversed, and all validation rules are calculated to determine the matching confidence of each validation feature pair with each rule. The matching confidence is calculated along two dimensions: first, fault type matching. If the fault type of the validation feature pair completely matches the fault type in the rule, this dimension scores 1; if partially related, a score between 0 and 1 is assigned based on the degree of correlation; if completely unrelated, the score is 0. Second, environmental data matching. This involves determining whether the current temperature and humidity fall within the range specified in the rule. If completely within the range, this dimension scores 1; if partially within the range, a score between 0 and 1 is assigned based on the overlap ratio; if completely outside the range, the score is 0. The scores from both dimensions are multiplied to obtain the matching confidence of the rule.

[0043] Extract the root cause pattern of the fault associated with the verification rule corresponding to the highest matching confidence, and use it as the result of the fault root cause determination for critical equipment.

[0044] After inferring the root cause of failure in critical equipment, the following process is also included: constructing a set of potential defect characteristics for the critical equipment. This set of potential defect characteristics includes excessively high environmental humidity, signs of component aging, and hereditary defects. Excessively high environmental humidity refers to the humidity of the equipment's operating environment exceeding the equipment's design tolerance for an extended period; signs of component aging refer to the declining trend in the performance parameters of key components over time; and hereditary defects refer to design or manufacturing defects that are prevalent in the same batch or model of equipment.

[0045] Obtain equipment attribute information for critical equipment. This information includes equipment model, production batch, service life, maintenance history, and component composition. The equipment model identifies the equipment's specifications and functions; the production batch records the production time and production line information; the service life reflects the equipment's operational duration; the maintenance history includes past repair records and replaced parts; and the component composition lists the equipment's main components and their models.

[0046] By combining equipment attribute information and environmental monitoring data, it is determined whether major equipment meets the potential defect characteristic set, and the occurrence probability of each potential defect characteristic is generated to form a risk identification list. When the ambient humidity is too high, real-time humidity data is compared with the equipment's design humidity tolerance, and the percentage of time the humidity exceeds the standard is statistically analyzed. When identifying signs of component aging, current performance parameters are compared with those of newer equipment, and the degree of aging is assessed in conjunction with the service life and maintenance records. When identifying inherited defects, fault records of the same batch or model of equipment are queried, and the occurrence ratio of similar defects is statistically analyzed.

[0047] The following is a reference data table for judging potential defect characteristics: High ambient humidity Real-time humidity value, design tolerance humidity Humidity sensor, equipment manual Calculate the proportion of time with excessive humidity to the total operating time. Signs of component aging Current performance parameters, new equipment parameters Performance monitoring system, equipment factory report Compare the rate of decrease in parameters and assess in conjunction with the service life. Family genetic defects Fault records of equipment in the same batch Equipment Fault Database Statistical analysis of the incidence rate of similar defects in the same batch of equipment. Based on the above assessment results, the probability of occurrence of each potential defect feature is generated, forming a risk identification list. The risk identification list is arranged from highest to lowest probability, clearly showing the likelihood of each potential defect. Early warning signals for critical equipment are generated based on the risk identification list. These signals are color-coded according to their probability of occurrence, with higher probabilities resulting in more prominent colors, facilitating timely attention and handling by relevant personnel.

[0048] Example 5: The probability of occurrence of trap features is used to form a risk identification list. The specific process is as follows: Equipment attribute information is decomposed, and multiple attribute feature sub-values ​​are extracted. Equipment attribute information includes equipment model, production batch, service life, maintenance history, and component composition. The equipment model is converted into a coded value, with different models corresponding to different codes. The coding rules are based on the equipment's functional category and specifications. The production batch is converted into a time series value based on the production time, accurate to the month. For example, if a piece of equipment was produced in March 2020, the corresponding value is 202003. The service life is quantified in months, directly taking the total number of months from when the equipment was put into use to the present. The maintenance history is represented by two sub-values: maintenance frequency and average maintenance interval. The maintenance frequency is the cumulative number of repairs to the equipment, and the average maintenance interval is the average number of months between two consecutive maintenance sessions. The component composition is converted into values ​​based on the importance of each component and supplier information. Core components are assigned higher base values, and components from different suppliers are adjusted based on past quality records.

[0049] The attribute feature sub-values ​​are input into a pre-trained risk assessment model to generate the probability of occurrence of each potential defect feature. The attribute feature sub-values ​​are then normalized to generate standardized feature values. During normalization, the historical maximum and minimum values ​​of each sub-value are first determined, and a linear transformation maps the original sub-values ​​to the range of 0 to 1. For equipment model codes, adjustments are made based on the historical failure frequency of that model; the higher the failure frequency, the closer the standardized feature value of the corresponding code is to 1. The standardized feature value of a production batch is positively correlated with the service life of that batch of equipment; the longer the service life, the higher the standardized feature value. The standardized feature value of service life directly reflects the degree of equipment aging; the closer the service life is to the upper limit of the design life, the closer the standardized feature value is to 1. The standardized feature value of the number of maintenance operations in the maintenance history is positively correlated with the maintenance frequency; the more frequent the maintenance, the higher the value. The standardized feature value of the average maintenance interval is negatively correlated with the interval length; the shorter the interval, the higher the value. The standardized feature value of component composition is set based on the component failure rate; components with higher failure rates correspond to higher values.

[0050] Based on standardized eigenvalues, the contribution weight of each potential defect feature is calculated. The contribution weight is determined by the correlation between each attribute feature sub-value and the occurrence of the potential defect, derived by analyzing the correspondence between attribute features and defect occurrences in historical data. For the potential defect of excessive environmental humidity, the attribute feature sub-values ​​related to environmental humidity (such as the proportion of maintenance records due to humidity in the maintenance history) have a higher contribution weight. For signs of component aging, the service life and maintenance frequency have greater contribution weights. For inherited defects, the production batch and equipment model have more prominent contribution weights.

[0051] By combining contribution weights and standardized eigenvalues, the probability of occurrence for each potential defect feature is generated. The probability of occurrence is obtained by summing the standardized eigenvalues ​​of each attribute feature sub-value multiplied by its corresponding contribution weight; the larger the sum, the higher the probability of occurrence for that potential defect feature. During the calculation, the combinations of attribute feature sub-values ​​differ for different potential defect features. For example, excessively high ambient humidity is mainly associated with humidity sensor data and relevant entries in maintenance records; signs of component aging are mainly associated with service life and performance monitoring data; and hereditary defects are mainly associated with production batches and failure data of the same model of equipment.

[0052] A risk identification list is constructed by summarizing the occurrence probabilities of various potential defect characteristics. The risk identification list is sorted by probability of occurrence, clearly presenting the risk level of each potential defect. Early warning signals for critical equipment are generated based on the risk identification list. These signals are presented in the form of text descriptions and icon labels, with different icon colors and text descriptions corresponding to different probability ranges, enabling relevant personnel to quickly identify potential risks to the equipment. The risk identification list and early warning signals are updated in real time, dynamically adjusting to changes in equipment attribute information and environmental monitoring data to ensure timely reflection of the potential defect status of the equipment.

[0053] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0054] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent diagnosis of major equipment faults driven by an edge diagnostic model algorithm, characterized in that, include: Collect real-time monitoring data streams during the operation of critical equipment. The real-time monitoring data streams include vibration signals, temperature readings, and acoustic characteristics. Generate equipment operating status waveform diagrams and equipment status diagnosis auxiliary documents based on the real-time monitoring data streams. The waveform diagram of the equipment's operating status and the auxiliary document for equipment status diagnosis are input into a pre-configured edge diagnosis model for processing to obtain initial classification data of the fault types corresponding to the critical equipment. Based on the waveform spectrum of the equipment's operating status, multiple fault mechanism feature parameters of the critical equipment are extracted and matched with a predefined set of fault mechanism models. Based on the result of the matching operation, auxiliary classification data corresponding to the fault type of the critical equipment is generated. By integrating the initial classification data and the auxiliary classification data, the final failure type of the critical equipment is determined, and the root cause of the failure of the critical equipment is inferred based on the final failure type.

2. The intelligent diagnostic method for major equipment faults driven by the edge diagnostic model algorithm according to claim 1, characterized in that, Based on the real-time monitoring data stream, generate equipment operating status waveform diagrams and equipment status diagnostic auxiliary documents, including: Based on real-time monitoring data stream, a waveform spectrum of equipment operation status is constructed, and multiple fault mechanism feature components are extracted from the real-time monitoring data stream. The fault mechanism feature components cover amplitude dispersion, phase concentration, polarity effect performance, flight trajectory characteristics, pulse balance, low-frequency component characteristics, and high-frequency component characteristics. The polarity effect is determined by analyzing the difference in energy distribution of the signal in the positive and negative half-cycles; the flight trajectory characteristics are obtained by plotting trajectory curves from continuously acquired displacement signals, and then analyzing the curvature changes and offset trends of the curves. Quantify the correlation between real-time monitoring data streams and environmental monitoring data to generate external impact factors of faults; Each fault mechanism feature component and external influencing factor is discretized and converted into text, and the processing results are represented in key-value pair form to form an auxiliary document for equipment status diagnosis.

3. The intelligent diagnostic method for major equipment faults driven by the edge diagnostic model algorithm according to claim 2, characterized in that, The environmental monitoring data includes real-time temperature data and real-time humidity data; The correlation between the quantified real-time monitoring data stream and the environmental monitoring data is used to generate external impact factors of the fault, including: Dynamic covariance calculation is performed on real-time monitoring data streams and real-time temperature data within the same time period to obtain the temperature correlation coefficient; Dynamic covariance calculation is performed on real-time monitoring data streams and real-time humidity data within the same time period to obtain the humidity correlation coefficient; The temperature correlation coefficient and humidity correlation coefficient are weighted and superimposed according to a preset weight ratio to output the external influence factor of the fault.

4. The intelligent diagnostic method for major equipment faults driven by the edge diagnostic model algorithm according to claim 1, characterized in that, The pre-configured edge diagnostic model includes a temporal coding module, a text parsing module, a multimodal fusion module, and a multimodal output module; The waveform graph of the equipment's operating status and the auxiliary document for equipment status diagnosis are input into a pre-configured edge diagnostic model for processing to obtain initial classification data for the fault types corresponding to the critical equipment, including: The timing coding module encodes the waveform of the device's operating status to generate waveform feature vectors. The text parsing module encodes the device status diagnostic auxiliary document to generate a text feature vector. Based on the cross-modal alignment loss function and the multimodal matching loss function, the waveform feature vector and the text feature vector are input into the multimodal fusion module for processing to obtain the multimodal fusion result; The multimodal fusion results are input into the multimodal output module for processing, and the initial classification data corresponding to the fault types of critical equipment is output.

5. The intelligent diagnostic method for major equipment faults driven by the edge diagnostic model algorithm according to claim 1, characterized in that, Based on the waveform spectrum of the equipment's operating status, multiple fault mechanism feature parameters of the critical equipment are extracted and matched with a predefined set of fault mechanism models. Based on the results of the matching operation, auxiliary classification data corresponding to the fault type of the critical equipment is generated, including: The waveform spectrum of the equipment's operating status is quantified by multiple fault mechanism characteristic parameters, including amplitude level, abnormal event count, event interval duration, comparison of low-frequency and high-frequency components, positive and negative half-axis symmetry characteristics, single-peak or double-peak performance, fractal characteristics, ultrasonic detection probability, and pulse sequence characteristics. Multiple fault mechanism feature parameters are matched with a set of predefined fault mechanism models corresponding to various fault types to generate multiple matching degree values. The fault type associated with the fault mechanism model corresponding to the highest matching degree value is used as auxiliary classification data.

6. The intelligent diagnostic method for major equipment faults driven by the edge diagnostic model algorithm according to claim 5, characterized in that, The initial classification data includes fault types and their corresponding probability values; By integrating the initial classification data and the auxiliary classification data, the final failure type of the critical equipment is determined, including: The weight factors for the initial classification data are calculated based on the probability values, and the auxiliary weight factors for the auxiliary classification data are calculated based on the matching degree values. The weighting factors and auxiliary weighting factors are normalized to generate normalized weighting factors; By combining normalized weighting factors, initial classification data, and auxiliary classification data, the final failure type of critical equipment is determined.

7. The intelligent diagnostic method for major equipment faults driven by the edge diagnostic model algorithm according to claim 1, characterized in that, Inferring the root causes of failures in critical equipment includes: Construct a cross-validation rule base, which contains the mapping relationship between fault modes recorded in the historical fault database and environmental monitoring data; The final fault type is combined with the current environmental monitoring data to form a verification feature pair; Traverse all validation rules in the cross-validation rule base and calculate the matching confidence of the validation feature pair with each validation rule; Extract the fault root cause pattern associated with the verification rule corresponding to the highest matching confidence, and use it as the fault root cause determination result of the critical equipment.

8. The intelligent diagnostic method for major equipment faults driven by the edge diagnostic model algorithm according to claim 1, characterized in that, After inferring the root cause of the failure of the critical equipment, the following is also included: Construct a set of potential defect features for critical equipment, including excessively high ambient humidity, signs of component aging, and familial genetic defects; Obtain equipment attribute information of major equipment, and combine the equipment attribute information with environmental monitoring data to determine whether the major equipment meets the potential defect feature set, and generate the probability of occurrence of each potential defect feature to form a risk identification list; Early warning signals for critical equipment are generated based on the risk identification list.

9. The intelligent diagnostic method for major equipment faults driven by the edge diagnostic model algorithm according to claim 8, characterized in that, The generation of the probability of occurrence of each potential defect feature to form a risk identification list includes: Decompose the device attribute information and extract multiple attribute feature sub-values; The attribute feature sub-values ​​are input into a pre-trained risk assessment model to generate the probability of occurrence of each potential defect feature; Summarize the probability of occurrence of each potential defect feature to construct a risk identification list.

10. The intelligent diagnostic method for major equipment faults driven by the edge diagnostic model algorithm according to claim 8, characterized in that, The attribute feature sub-values ​​are input into a pre-trained risk assessment model to generate the occurrence probability of each potential defect feature, including: Normalize the attribute feature sub-values ​​to generate standardized feature values; Based on standardized feature values, the contribution weight of each potential defect feature is calculated; By combining contribution weights and standardized feature values, the probability of occurrence of each potential defect feature is generated.

Images