Intelligent equipment fault diagnosis method and system based on Modbus protocol

Abstract

The invention relates to the technical field of equipment fault intelligent diagnosis, in particular to an equipment fault intelligent diagnosis method and system based on a Modbus protocol. The method comprises the following steps: acquiring real-time operation data from target industrial equipment through a Modbus protocol, dynamically adjusting an initial sampling frequency based on an equipment operation state, and performing multiple verification and compensation correction on the acquired data to obtain a stable data stream; performing multi-scale decomposition and feature enhancement processing on the stable data stream, extracting a time-frequency domain mixed feature set, and constructing a feature evolution trajectory; inputting the feature evolution trajectory into a double-branch diagnosis model integrating equipment state prediction and fault classification, and outputting an equipment health degree score and fault type probability distribution; and constructing a dynamic fault threshold curved surface, carrying out multi-dimensional fusion decision by combining the equipment health degree score and the fault type probability distribution, and generating a graded fault early warning and maintenance strategy. According to the invention, the accuracy, timeliness and adaptability of industrial equipment fault diagnosis can be greatly improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent equipment fault diagnosis technology, and in particular to an intelligent equipment fault diagnosis method and system based on the Modbus protocol. Background Technology

[0002] In the field of industrial equipment fault diagnosis, early diagnostic methods relied on manual inspections, which were highly subjective, difficult to detect early faults, and inefficient. Later, diagnostic technologies based on sensors and the Modbus protocol were developed, which collected data through fixed sampling frequencies and combined threshold judgments, simple trend analysis, or single-function machine learning models for diagnosis. However, this technology still has significant problems, mainly manifested in the fact that fixed sampling frequencies easily lead to data redundancy or missing key data, and data verification and correction are simple but of low quality. Furthermore, fault feature extraction is singular and lacks optimization; the diagnostic model has limited functionality, poor adaptability to operating environments, and is difficult to meet the requirements for stable equipment operation. Summary of the Invention

[0003] This invention addresses the technical problems of poor accuracy, timeliness, and adaptability in the fault diagnosis of industrial equipment in the prior art by providing a method and system for intelligent equipment fault diagnosis based on the Modbus protocol.

[0004] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: In a first aspect, the present invention provides an intelligent equipment fault diagnosis method based on the Modbus protocol, comprising: collecting real-time operating data from a target industrial equipment via the Modbus protocol; dynamically adjusting the initial sampling frequency based on the equipment operating status; and performing multiple verifications and compensation corrections on the collected data to obtain a stable data stream; performing multi-scale decomposition and feature enhancement processing on the stable data stream to extract a time-frequency domain hybrid feature set and constructing a feature evolution trajectory; inputting the feature evolution trajectory into a dual-branch diagnostic model integrating equipment status prediction and fault classification, and outputting an equipment health score and a fault type probability distribution; constructing a dynamic fault threshold surface based on historical equipment operating data and equipment operating environment parameter data; and performing multi-dimensional fusion decision-making by combining the equipment health score and the fault type probability distribution to generate a graded fault warning and maintenance strategy.

[0005] Optionally, real-time operating data is collected from the target industrial equipment via the Modbus protocol. The initial sampling frequency is dynamically adjusted based on the equipment's operating status, and the collected data undergoes multiple verifications and compensation corrections to obtain a stable data stream. This includes: configuring Modbus communication parameters, wherein the Modbus communication parameters include the device address, register address, and initial sampling frequency; monitoring the data fluctuation characteristics of the target industrial equipment's current, voltage, and temperature, and adaptively adjusting the initial sampling frequency according to the fluctuation amplitude; performing missing data detection, jump correction, and noise filtering on the collected data, and using a sliding window compensation mechanism to repair abnormal sections to form the stable data stream.

[0006] Optionally, the stable data stream is subjected to multi-scale decomposition and feature enhancement processing to extract a time-frequency domain hybrid feature set and construct a feature evolution trajectory, including: performing wavelet packet multi-scale decomposition on the stable data stream to obtain different frequency band components; extracting the time-domain statistical features and frequency-domain energy features of each component to construct an initial feature set; screening key features based on grey relational analysis and enhancing feature distinguishability through feature transformation to generate the time-frequency domain hybrid feature set; and constructing a feature evolution trajectory by combining the time-frequency domain hybrid feature set.

[0007] The method of constructing a feature evolution trajectory by combining the time-frequency domain hybrid feature set includes: sorting the time-frequency domain hybrid feature set according to the acquisition time order to obtain a feature sequence, and segmenting the feature sequence using a sliding window of fixed length; calculating the mean, standard deviation, and range of feature values ​​within each sliding window to form window feature statistics; analyzing the morphological similarity and feature change trend of feature sequences between adjacent sliding windows based on a dynamic time warping algorithm; and extracting the peak points, valley points, and inflection points of the feature sequence to construct a feature evolution key point sequence, thus forming a feature evolution trajectory of the device status.

[0008] Optionally, the construction steps of the dual-branch diagnostic model include: collecting a set of sample feature evolution trajectories based on historical equipment operation data, and labeling the equipment status corresponding to each feature evolution trajectory to obtain a set of sample equipment statuses, wherein the sample equipment statuses include equipment health status level identifiers and fault type identifiers; constructing the network architecture of the dual-branch diagnostic model based on machine learning, including an equipment status prediction branch and an equipment fault classification branch; and using the set of sample feature evolution trajectories and the set of sample equipment statuses to perform supervised training on the dual-branch diagnostic model until the verification accuracy converges, thus completing the model construction.

[0009] Optionally, the construction steps of the dynamic fault threshold surface include: collecting a set of feature evolution trajectories under multiple operating conditions based on historical equipment operating data, and labeling the equipment state corresponding to each feature evolution trajectory to obtain a set of sample equipment states; dividing the feature evolution trajectory set into operating conditions based on the K-means clustering algorithm to determine typical operating condition categories; constructing a benchmark feature library for each typical operating condition and calculating the confidence interval of feature values ​​under each operating condition; establishing a dynamic fault threshold surface based on the multi-operating condition benchmark feature library, and introducing equipment operating environment parameter data to correct the dynamic fault threshold surface in real time.

[0010] Specifically, a dynamic fault threshold surface is established based on a multi-condition benchmark feature library, and equipment operating environment parameter data is introduced to correct the dynamic fault threshold surface in real time. This includes: collecting equipment operating environment parameter data and constructing an environmental parameter dataset, wherein the equipment operating environment parameter data includes ambient temperature, ambient humidity, and equipment load rate; mapping confidence intervals in each operating condition benchmark feature library to the feature-environment parameter space based on a multi-dimensional spatial interpolation algorithm to construct an initial fault threshold surface; introducing an environmental parameter correction factor and using weighted calculation to dynamically adjust the initial fault threshold surface to obtain a corrected dynamic fault threshold surface; and continuously updating the dynamic fault threshold surface through a sliding window mechanism based on real-time collected equipment operating environment parameters and equipment operating data.

[0011] Optionally, a multi-dimensional fusion decision is made by combining the equipment health score and the fault type probability distribution to generate a graded fault warning and maintenance strategy. This includes: normalizing the equipment health score and the fault type probability distribution through weighted fusion calculation to obtain a comprehensive risk value; performing condition-adaptive calibration on the comprehensive risk value by referring to the threshold range in the dynamic fault threshold surface that matches the current environmental conditions; classifying warning levels based on the calibrated comprehensive risk value, wherein the warning levels include normal level, attention level, warning level, and emergency level; querying the equipment maintenance knowledge base based on the warning level, matching and generating corresponding maintenance strategies, and determining maintenance priorities.

[0012] Secondly, the present invention provides an intelligent equipment fault diagnosis system based on the Modbus protocol, comprising: The real-time running data acquisition module is used to acquire real-time running data from the target industrial equipment via the Modbus protocol, dynamically adjust the initial sampling frequency based on the equipment's operating status, and perform multiple verifications and compensation corrections on the acquired data to obtain a stable data stream. The feature evolution trajectory construction module is used to perform multi-scale decomposition and feature enhancement processing on the stable data stream, extract the time-frequency domain hybrid feature set, and construct the feature evolution trajectory. The dual-branch diagnostic module is used to input the feature evolution trajectory into the dual-branch diagnostic model that integrates equipment status prediction and fault classification, and output the equipment health score and fault type probability distribution. The multi-dimensional fusion decision module is used to construct a dynamic fault threshold surface based on the equipment's historical operating data and equipment operating environment parameter data. It combines the equipment health score and fault type probability distribution to perform multi-dimensional fusion decision-making and generate graded fault early warning and maintenance strategies.

[0013] By implementing this invention, it is possible to collect real-time operating data from target industrial equipment via the Modbus protocol, dynamically adjust the initial sampling frequency based on the equipment's operating status, and perform multiple verifications and compensation corrections on the collected data to obtain a stable data stream. The multiple verification and compensation correction operations can effectively eliminate errors, noise, and missing parts in the collected data, ensuring the accuracy and reliability of the data used for subsequent analysis, and laying a good data foundation for subsequent fault diagnosis. By adopting the widely used Modbus protocol, this diagnostic method is compatible with many industrial devices that support the protocol, thus expanding the applicability of the method. By implementing this invention, multi-scale decomposition and feature enhancement processing can be performed on the stable data stream to extract a time-frequency domain hybrid feature set and construct a feature evolution trajectory. Multi-scale decomposition combined with time-frequency domain feature extraction can mine feature information in equipment operation data from different angles and levels, avoiding the limitations of a single feature extraction method, and more comprehensively reflecting the equipment operation status. The feature evolution trajectory can clearly show the changing trend of equipment features over time, making it easier for staff to intuitively understand the evolution process of equipment operation status and providing an intuitive basis for equipment status prediction and fault diagnosis. By implementing this invention, the feature evolution trajectory can be input into a dual-branch diagnostic model that integrates equipment status prediction and fault classification, and the output can be an equipment health score and a fault type probability distribution. The dual-branch diagnostic model has both equipment status prediction and fault classification functions, and can output relevant information on equipment health status and possible fault types at once, improving the comprehensiveness and efficiency of fault diagnosis and reducing the cumbersome process of performing status prediction and fault classification separately. The equipment health score and fault type probability distribution output by the dual-branch diagnostic model provide quantitative diagnostic basis for staff, making it easier for staff to more accurately assess equipment fault conditions and make reasonable decisions. By implementing this invention, a dynamic fault threshold surface can be constructed based on historical equipment operating data and equipment operating environment parameter data. This surface, combined with the equipment health score and fault type probability distribution, enables multi-dimensional fusion decision-making, generating tiered fault warning and maintenance strategies. The dynamic fault threshold surface can be adjusted according to different equipment operating conditions and environmental parameters, avoiding the problem of inaccurate judgment under complex and changing conditions caused by fixed thresholds. This improves the adaptability and accuracy of fault judgment. Furthermore, by matching maintenance strategies with warning levels and determining maintenance priorities, maintenance work becomes more targeted and planned, allowing for the rational allocation of maintenance resources, reducing maintenance costs, and improving equipment availability and lifespan.

[0014] In summary, by implementing this invention, the accuracy, timeliness, and adaptability of industrial equipment fault diagnosis can be significantly improved, the misdiagnosis and missed diagnosis rates of equipment faults can be effectively reduced, unnecessary maintenance costs can be reduced, and a strong guarantee can be provided for the continuous and stable operation of industrial production. Attached Figure Description

[0015] Figure 1 A flowchart illustrating an intelligent device fault diagnosis method based on the Modbus protocol provided by this invention; Figure 2 This is a schematic diagram of the structure of an intelligent equipment fault diagnosis system based on the Modbus protocol provided by the present invention.

[0016] In the attached diagram, the components represented by each number are as follows: The system includes a real-time data acquisition module 11, a feature evolution trajectory construction module 12, a dual-branch diagnosis module 13, and a multi-dimensional fusion decision module 14. Detailed Implementation

[0017] 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.

[0018] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0019] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0020] Example 1, as Figure 1 As shown, this embodiment of the invention provides a device fault intelligent diagnosis method based on the Modbus protocol, including: S100: Collects real-time operating data from target industrial equipment via the Modbus protocol, dynamically adjusts the initial sampling frequency based on the equipment's operating status, and performs multiple verifications and compensation corrections on the collected data to obtain a stable data stream; S200: Perform multi-scale decomposition and feature enhancement processing on the stable data stream, extract the time-frequency domain hybrid feature set, and construct the feature evolution trajectory; S300: Input the feature evolution trajectory into the dual-branch diagnostic model that integrates equipment status prediction and fault classification, and output the equipment health score and fault type probability distribution; S400: Based on historical equipment operation data and equipment operating environment parameter data, a dynamic fault threshold surface is constructed. The equipment health score and fault type probability distribution are combined to perform multi-dimensional fusion decision-making and generate graded fault early warning and maintenance strategies.

[0021] In step S100 of this application embodiment, real-time operating data is collected from the target industrial equipment via the Modbus protocol. The initial sampling frequency is dynamically adjusted based on the equipment's operating status, and the collected data undergoes multiple verifications and compensation corrections to obtain a stable data stream, including: Configure Modbus communication parameters, which include device address, register address, and initial sampling frequency; Monitor the data fluctuation characteristics of the target industrial equipment's current, voltage, and temperature, and adaptively adjust the initial sampling frequency based on the fluctuation amplitude; The collected data is subjected to missing data detection, jump correction and noise filtering, and an abnormal segment is repaired by a sliding window compensation mechanism to form the stable data stream.

[0022] In this embodiment of the application, the purpose of step S100 is to provide high-quality data support for subsequent feature extraction and fault diagnosis. By configuring key parameters of the Modbus protocol, a stable communication link between the diagnostic system and the target industrial equipment is established to avoid data acquisition failure or data corruption caused by protocol parameter mismatch. Furthermore, through multiple verification and compensation corrections, interference such as missing data, jumps, and noise in the acquired data is eliminated, and a "stable data stream" is output to avoid inferior data causing deviations in subsequent feature extraction or misjudgments in fault diagnosis.

[0023] To achieve the above objectives, Modbus communication parameters must first be configured. These parameters include the device address, register address, and initial sampling frequency. Specifically, you need to consult the Modbus protocol manual of the target industrial device to confirm the supported Modbus type (e.g., RTU / TCP); the register addresses corresponding to key operational data (e.g., voltage stored in register 0x0001, current stored in register 0x0002); and the device address range (e.g., 1-247).

[0024] Then, enter the above parameters into the real-time operation data acquisition module of the diagnostic system and start the communication test—send a Modbus read command to the target industrial equipment, such as function code 0x03 to read the holding register. If the correct initial operation data can be received and parsed, the parameter configuration is complete; if the communication fails, check for device address conflicts, register address errors, or protocol type mismatches until a connection is established.

[0025] Furthermore, it is necessary to monitor the data fluctuation characteristics of the target industrial equipment's current, voltage, and temperature, and adaptively adjust the initial sampling frequency based on the fluctuation amplitude. Specifically, this requires real-time acquisition of the target industrial equipment's current, voltage, and temperature data via a configured Modbus link, calculating the "fluctuation amplitude" per unit time—that is, the difference between the maximum and minimum values ​​of a certain parameter. For example, if the voltage has a maximum value of 382V and a minimum value of 378V within 10 seconds, the fluctuation amplitude is 4V.

[0026] Then, it is necessary to preset the correspondence rules between fluctuation amplitude and sampling frequency. For example, if the fluctuation amplitude of any parameter is less than or equal to the preset threshold, such as voltage fluctuation ≤ 2V, current fluctuation ≤ 5A, and temperature fluctuation ≤ 1℃, then the device is considered to be in a stable state, and the sampling frequency is reduced to 50% of the initial frequency, such as from 1Hz to 0.5Hz; if the fluctuation amplitude of any parameter is greater than the preset threshold, then the device is considered to be in a large state fluctuation, and the sampling frequency is increased to 2-3 times the initial frequency, such as from 1Hz to 2Hz; thereafter, according to a predetermined time interval, such as 10s, the fluctuation monitoring and frequency adjustment process is repeated to ensure that the sampling frequency always matches the current state of the device.

[0027] Furthermore, it is necessary to perform missing data detection, jump correction, and noise filtering on the collected data, and to use a sliding window compensation mechanism to repair abnormal sections, thereby forming the stable data stream.

[0028] The missing data detection process involves collecting data in a time-series manner. If there is no data at a certain time point or the data is marked as "invalid", it is determined that the data is missing. Then, a "sliding window compensation mechanism" is used to fill the missing data. That is, taking the missing point as the center, taking N valid data points before and after it, calculating the mean of these data, and using the mean to fill the missing point. For example, when N=5, the five voltage values ​​before the missing point are taken as 379, 380, 381, 380, and 379, with a mean of 380, and 380 is used to fill the missing point.

[0029] Jump correction involves calculating the difference between two adjacent sampling points. If the difference exceeds a preset jump threshold, the subsequent data point is considered an anomaly. For example, the voltage jump threshold is 20V, and the current jump threshold is 50A. Then, the anomaly point is discarded, and a sliding window compensation mechanism is used to replace it. For instance, the three voltages preceding the jump point (380, 381, 380) and the last two (381, 380) are taken, with an average of 380.2. This 380.2 value is then used to replace the 500V jump point.

[0030] Noise filtering can be performed using either wavelet threshold filtering or moving average filtering. Taking moving average filtering as an example, a window is formed by taking M points before and after each data point, calculating the mean of the data within the window, and replacing the original data point with the mean. For example, if the original temperature data is 50.2, 49.6, 50.1, 49.8, and 50.3, when the window M=3, the mean of the second point is (50.2+49.6+50.1) / 3≈49.97, which replaces the original 49.6, thus eliminating random noise.

[0031] After completing the above three steps, traverse all data points to confirm that there are no missing data points, no jumps, and low noise, forming a "stable data stream," and then pass the stable data stream to the next step.

[0032] In step S200 of this application embodiment, the stable data stream undergoes multi-scale decomposition and feature enhancement processing to extract a time-frequency domain hybrid feature set and construct a feature evolution trajectory, including: The stable data stream is subjected to wavelet packet multi-scale decomposition to obtain components of different frequency bands; Extract the time-domain statistical features and frequency-domain energy features of each component to construct an initial feature set; Key features are selected based on grey relational analysis, and feature differentiation is enhanced through feature transformation to generate the time-frequency domain hybrid feature set. The feature evolution trajectory is constructed by combining the time-frequency domain hybrid feature set analysis.

[0033] In this embodiment of the application, the purpose of step S200 is to transform the "stable data stream" into a "feature evolution trajectory" that can reflect changes in the device state through multi-scale decomposition, feature extraction, filtering enhancement and trajectory construction, so as to provide input for the subsequent dual-branch diagnostic model.

[0034] To achieve the above objectives, the stable data stream first needs to be decomposed using wavelet packets at multiple scales to obtain components in different frequency bands. Specifically, the appropriate wavelet packet basis function needs to be selected based on the type of the stable data stream. For example, in a stable data stream, current is a periodic signal, and temperature is a trending signal; therefore, wavelet packet basis functions such as the db4 wavelet packet and the sym8 wavelet packet can be used. For instance, the db4 wavelet packet performs better in decomposing abrupt signals, so it can be selected for current signals.

[0035] Then, based on the diagnostic accuracy requirements and data sampling frequency, the number of wavelet packet decomposition layers is determined, typically 3-5 layers. For example, temperature data with a sampling frequency of 2Hz can be decomposed into 8 frequency band components by 3 layers, covering the frequency range from 0-0.25Hz (low frequency) to 1.75-2Hz (high frequency).

[0036] Next, the stable data stream is decomposed layer by layer using the wavelet packet decomposition algorithm. Each layer of decomposition further breaks down each frequency band component of the previous layer into "approximate components" and "detail components", and finally obtains the frequency band components of all decomposition layers. For example, after 3 layers of decomposition, 8 frequency band components are obtained, and the time series data of each component are stored.

[0037] Furthermore, it is necessary to extract the time-domain statistical features and frequency-domain energy features of each component to construct an initial feature set. For example, five types of core time-domain statistical features can be calculated for the time series of each frequency band component, including central tendency features, dispersion features, morphological features, trend features, and periodic features.

[0038] Among them, the central tendency features are mainly the mean and median; the dispersion features are mainly the variance, standard deviation, and range; the morphological features are mainly the peak factor, which is the ratio of the peak value to the effective value, used to capture instantaneous spike signals, such as a short circuit in equipment may cause the peak factor to rise sharply, and the impulse factor, which is the ratio of the peak value to the mean; the trend features are mainly the linear fitting slope, used to reflect the changing trend of the components, such as a positive and increasing slope of a low-frequency temperature component, which may indicate a heat dissipation failure in the equipment; the periodic features are mainly the autocorrelation coefficient, which reflects the periodicity of the components, such as a phase loss in a motor may cause an abnormal periodicity in the autocorrelation coefficient of the current component.

[0039] Next, a Fourier transform is performed on each frequency band component to convert the time-domain signal into a frequency-domain signal. Then, three types of core frequency-domain energy features are extracted: frequency band energy value, energy proportion, and energy entropy. The frequency band energy value is the total energy of the component within its corresponding frequency range; the energy proportion is the ratio of the component's energy value to the total energy of all frequency band components; and the energy entropy is the information entropy of the energy of each frequency band component, reflecting the uniformity of energy distribution. During equipment failure, the energy distribution may change from uniform to concentrated, leading to a decrease in energy entropy.

[0040] Then, the "time domain statistical features" and "frequency domain energy features" of all frequency band components are summarized to form an initial feature set. For example, for 8 frequency band components, 10 time domain features and 3 frequency domain features are extracted for each component, and the initial feature set has a total of 104 features.

[0041] Furthermore, it is necessary to screen key features based on grey relational analysis and enhance feature discrimination through feature transformation to generate the time-frequency domain hybrid feature set.

[0042] First, "equipment status labels" are used as a reference sequence to reflect the actual status of the equipment, such as healthy = 0, minor fault = 1, and severe fault = 2. Then, each feature in the initial feature set is used as a comparison sequence. Grey relational analysis is used to calculate the "grey relational degree" between each comparison sequence and the reference sequence—the closer the relational degree is to 1, the stronger the correlation between the feature and the equipment status. A relational degree threshold is set, such as 0.6. Features with a relational degree ≥ the threshold are retained, while redundant features with a relational degree < the threshold are removed, resulting in a filtered subset of key features.

[0043] Optionally, for the filtered subset of key features, a two-type feature transformation method can be used to enhance discriminative power: The first is standardization transformation, which involves performing Z-score standardization (feature value - mean) / standard deviation on each feature to eliminate the influence of "dimension differences" between different features. For example, the unit of current feature is A, and the unit of temperature feature is ℃. After standardization, the order of magnitude can be unified.

[0044] Second, nonlinear transformations are used. For features with weak distinguishability, logarithmic transformations, power transformations (such as squaring or square rooting), or principal component analysis (PCA) are employed. For example, performing a logarithmic transformation on the energy characteristics of a certain frequency band can amplify the differences between healthy and faulty states, thereby enhancing distinguishability.

[0045] Finally, the selected key features are merged with the transformed enhanced features to form the final time-frequency domain hybrid feature set. For example, 40 of the initial 104 features are retained after selection, and 10 enhanced features are added after transformation, resulting in a final hybrid feature set containing 50 features.

[0046] In step S200 of this application embodiment, constructing the feature evolution trajectory by combining the time-frequency domain hybrid feature set analysis includes: The time-frequency domain hybrid feature set is sorted according to the acquisition time order to obtain a feature sequence, and the feature sequence is segmented using a sliding window of fixed length; Within each sliding window, the mean, standard deviation, and range of the feature values ​​are calculated to form window feature statistics; Based on the dynamic time warping algorithm, the morphological similarity and feature change trend of feature sequences between adjacent sliding windows are analyzed. The peak points, valley points, and inflection points of the feature sequence are extracted to construct a key point sequence of feature evolution, forming the feature evolution trajectory of the equipment state.

[0047] To achieve the sorting of the time-frequency domain hybrid feature set according to the acquisition time order to obtain a feature sequence, and to segment the feature sequence using a fixed-length sliding window, the feature sequence first needs to be sorted. This involves arranging each feature in the time-frequency domain hybrid feature set according to the data acquisition time order, forming a "time series" for each individual feature. For example, the time series for the peak factor feature might be: [1.2, 1.3, 1.5, ..., 3.5, 4.0]. Then, a fixed-length sliding window, such as one with a window length of 10 and a step size of 1, is used to segment the time series of each feature. For instance, if the time series has 100 data points, after segmentation, 91 sliding windows are obtained, each containing 10 consecutive data points.

[0048] Furthermore, within each sliding window, the mean, standard deviation, and range of the feature values ​​need to be calculated to form window feature statistics. The mean reflects the average level of the feature within the window; for example, a mean peak factor of 2.5 within a window reflects the average fluctuation intensity during that time period. The standard deviation reflects the stability of the feature's fluctuations within the window; for example, a standard deviation of 0.3 indicates that the feature changes smoothly during that time period. The range reflects the maximum amplitude of the feature's changes within the window; for example, a range of 0.8 indicates that the maximum difference in the feature is relatively small during that time period.

[0049] Furthermore, based on the dynamic time warping algorithm, it is necessary to analyze the morphological similarity and feature change trends of feature sequences between adjacent sliding windows. The peak points, valley points, and inflection points of the feature sequences are extracted to construct a key point sequence of feature evolution, forming the feature evolution trajectory of the device state.

[0050] For trend analysis, the morphological similarity of "window feature statistics" between adjacent sliding windows can be calculated based on the Dynamic Time Warping (DTW) algorithm. Low similarity indicates a significant change in the feature state, such as a sudden increase from the mean of 2.5 to 3.8, which reduces the DTW similarity and can be marked as a "trend inflection point".

[0051] Next, the time series of all features are traversed to identify peaks, troughs, and inflection points. A "peak" is a local maximum value, a "trough" is a local minimum value, and an "inflection point" is a point where the trend direction changes, such as when the temperature slope changes from 0.1 to 0.5, forming a sequence of key points in the feature evolution.

[0052] Finally, the window feature statistics sequence of each feature is fused with the key point sequence, and a continuous curve is plotted along the time axis. For example, with time as the horizontal axis and the window mean of the peak factor as the vertical axis, peak points, valley points, and inflection points are marked to form the evolution trajectory of the feature. At the same time, the evolution trajectories of all key features are summarized to form a set of feature evolution trajectories for the entire device, which is then passed to the next step.

[0053] S300: Input the feature evolution trajectory into the dual-branch diagnostic model that integrates equipment status prediction and fault classification, and output the equipment health score and fault type probability distribution; In this embodiment, step S300 is the core decision-making step of the intelligent equipment fault diagnosis method. It achieves the dual goals of "dynamic prediction" and "accurate classification" of equipment status through a dual-branch diagnostic model, providing a quantitative basis for the operation and maintenance decisions of industrial equipment.

[0054] The construction steps of the dual-branch diagnostic model include: Based on historical equipment operation data, a set of sample feature evolution trajectories is collected, and the equipment status corresponding to each feature evolution trajectory is labeled to obtain a set of sample equipment statuses. The sample equipment statuses include equipment health status level identifiers and fault type identifiers. Based on machine learning, a network architecture for a dual-branch diagnostic model is constructed, including a device status prediction branch and a device fault classification branch. The dual-branch diagnostic model is trained under supervised supervision using the set of sample feature evolution trajectories and the set of sample device states until the verification accuracy converges, thus completing the model construction.

[0055] First, based on the historical operating data of the equipment, a set of sample feature evolution trajectories needs to be collected, and the equipment state corresponding to each feature evolution trajectory needs to be labeled to obtain a set of sample equipment states. Specifically, historical operating data of the target industrial equipment is collected, such as current, voltage, and temperature data from the past 3 years. Following steps S100 and S200, a set of sample feature evolution trajectories is generated, with each trajectory corresponding to the equipment state changes over a period of time.

[0056] Then, based on the actual operation records of the equipment, each trajectory is labeled with 5 equipment health status levels, such as Level 1 = Healthy, Level 2 = Slight Degradation, Level 3 = Moderate Degradation, Level 4 = Severe Degradation, and Level 5 = Fault. These can be subsequently converted into a health score of 0-100, such as Level 1 corresponding to 90-100 points, Level 5 corresponding to 0-20 points, and so on. If the trajectory corresponds to a fault status, the specific fault type is labeled, such as "bearing wear" or "motor phase loss," and stored using a one-hot encoding format, such as bearing wear corresponding to [1,0,0,...], and motor phase loss corresponding to [0,1,0,...].

[0057] Furthermore, it is necessary to construct a network architecture for a dual-branch diagnostic model based on machine learning, including a device status prediction branch and a device fault classification branch.

[0058] Firstly, temporal feature extraction networks such as LSTM and Transformer can be used as shared layers. Inputting the feature evolution trajectory allows for the extraction of key patterns within the trajectory, such as trend changes and the distribution of abrupt change points. For example, a three-layer LSTM network with 64 neurons per layer can be used to capture the long-term and short-term dependencies of the trajectory through a gating mechanism, such as the correlation between a small change in a feature one hour ago and the current fault.

[0059] For the device status prediction branch, the task objective is to regress and predict the device health score. After the output of the shared feature layer, two fully connected network layers are connected with 128 and 64 neurons respectively. The ReLU activation function is used, and finally, a health score of 0-100 is output through one neuron. The loss function is the mean squared error (MSE), which measures the difference between the predicted health score and the actual labeled level.

[0060] For the equipment fault classification branch, the task objective is to predict fault types in multiple categories. After the output of the shared feature layer, two fully connected networks are connected with 128 and 64 neurons respectively, using the ReLU activation function. Finally, the probability distribution of each category is output through neurons equal to the number of fault types. For example, if there are 5 fault types, then 5 neurons are configured. The cross-entropy loss function is used to measure the difference between the predicted probability distribution and the actual fault type.

[0061] Among them, the parameters of the shared feature layer are updated together in the training of the two branches to improve the efficiency of feature reuse; the total loss function is "device status prediction branch loss + equipment fault classification branch loss", and the importance of the two tasks can be adjusted by weighting, such as 0.6 for equipment status prediction and 0.4 for equipment fault classification.

[0062] Furthermore, the dual-branch diagnostic model needs to be trained under supervised supervision using the set of sample feature evolution trajectories and the set of sample device states until the verification accuracy converges.

[0063] Specifically, the set of sample feature evolution trajectories needs to be divided into training set, validation set, and test set in a 7:2:1 ratio; Ensure that the distribution of equipment health status levels and fault types is consistent in each set after partitioning. For example, the proportion of "bearing wear" samples should be 30% in both the training set and the validation set to avoid data distribution bias affecting the model's generalization ability.

[0064] For training the dual-branch diagnostic model, the optimizer used is the Adam optimizer with an initial learning rate of 0.001, which is dynamically adjusted based on the performance of the validation set. The training set is input into the model in batches of 32 trajectories, the total loss is calculated, and the parameters are updated via backpropagation. If the overall accuracy of the validation set, i.e., the accuracy of the MAE of the equipment status prediction branch and the accuracy of the equipment fault classification branch, does not improve for 5 consecutive epochs, training is stopped to avoid overfitting.

[0065] When the health prediction MAE score on the validation set is ≤5 and the fault classification accuracy is ≥90%, the model is considered to have converged. The model parameters at the time of convergence are saved as the final two-branch diagnostic model.

[0066] Finally, the feature evolution trajectory is input into a dual-branch diagnostic model that integrates equipment status prediction and fault classification, which can output the equipment health score and fault type probability distribution.

[0067] S400: Based on historical equipment operation data and equipment operating environment parameter data, a dynamic fault threshold surface is constructed. The equipment health score and fault type probability distribution are combined to perform multi-dimensional fusion decision-making and generate graded fault early warning and maintenance strategies.

[0068] In this embodiment, step S400 is the decision output link of the intelligent equipment fault diagnosis method. The core objective is to solve the problem that static thresholds are not suitable for dynamic working conditions in industrial scenarios. By constructing a dynamic standard for dynamic fault judgment, namely the dynamic fault threshold surface, and combining the equipment health score and fault type probability distribution output by S300, accurate hierarchical early warning and maintenance strategies are generated, realizing the transformation from diagnosis results to operation and maintenance actions.

[0069] The steps for constructing the dynamic fault threshold surface include: Based on the historical operating data of the equipment, a set of characteristic evolution trajectories under multiple working conditions is collected, and the equipment status corresponding to each characteristic evolution trajectory is labeled to obtain a sample set of equipment status. Based on the K-means clustering algorithm, the set of feature evolution trajectories is divided into working conditions to determine the typical operating condition categories. For each typical operating condition, a benchmark feature library is constructed, and the confidence interval of the feature values ​​under each condition is calculated. Based on a multi-condition benchmark feature library, a dynamic fault threshold surface is established, and equipment operating environment parameter data is introduced to correct the dynamic fault threshold surface in real time.

[0070] First, based on historical equipment operating data, it is necessary to collect a set of characteristic evolution trajectories under multiple operating conditions, and label the equipment state corresponding to each characteristic evolution trajectory to obtain a sample set of equipment states. Specifically, it is necessary to collect historical operating data of the equipment under different loads, different production rhythms, and different environmental conditions; then, process it according to the S100-S200 process to generate a set of characteristic evolution trajectories corresponding to different operating conditions, such as the "100% load + 35℃ environment" trajectory, the "50% load + 25℃ environment" trajectory, etc.

[0071] Then, each trajectory is labeled with a "normal / abnormal" status. Unlike the detailed level labeling of S300, this focuses on basic status judgment. Among them, the abnormal status needs to record the operating condition information at the same time, such as "abnormal bearing temperature under 100% load", to ensure that the real abnormality can be distinguished from the normal fluctuation under different operating conditions in the future.

[0072] Furthermore, based on the K-means clustering algorithm, the set of feature evolution trajectories needs to be divided into operating conditions to determine typical operating condition categories. First, core parameters that can characterize the operating conditions are selected, such as load rate, ambient temperature, and running time, to form an operating condition feature vector, such as [80% load, 30℃, continuous operation for 4 hours]. Then, for each feature evolution trajectory, its corresponding operating condition feature vector is associated.

[0073] Next, all operating condition feature vectors are input into the K-means algorithm, and operating conditions with high similarity are clustered into one class by calculating Euclidean distance. The number of clusters K is determined according to the actual operating scenario of the equipment. For example, K=5 corresponds to 5 typical operating conditions: no load-normal temperature, low load-normal temperature, high load-normal temperature, low load-high temperature, and high load-high temperature. The center vector of each typical operating condition is output. For example, the center vector of the "high load-high temperature" class is [90% load, 35℃, continuous operation for 6 hours], which serves as the benchmark for the subsequent construction of dynamic fault threshold.

[0074] Furthermore, it is necessary to construct a benchmark feature library for each typical operating condition and calculate the confidence interval of the feature values ​​under each condition. Specifically, for each type of typical operating condition, the feature evolution trajectory marked as "normal state" needs to be selected; all feature values ​​in these trajectories, such as time-domain statistical features and frequency-domain energy features, need to be extracted and stored according to feature type to form the benchmark feature library for that operating condition. For example, the peak factor feature library for the "high load-high temperature" operating condition contains 1000 peak factor values ​​under normal conditions.

[0075] Then, for each feature in the benchmark feature library, the 99.7% confidence interval is calculated using the 3σ principle. The lower limit of the interval is calculated as the feature mean minus 3 × standard deviation, and the upper limit is calculated as the feature mean plus 3 × standard deviation.

[0076] For example, if the mean of the current peak factor under the "high load-high temperature" condition is 3.2 and the standard deviation is 0.5, then the confidence interval is [3.2-1.5, 3.2+1.5]=[1.7, 4.7]. That is, the peak factor in this condition can be considered normal if it is in the range of 1.7-4.7.

[0077] Furthermore, it is necessary to establish a dynamic fault threshold surface based on a multi-condition benchmark feature library, and to introduce equipment operating environment parameter data to correct the dynamic fault threshold surface in real time.

[0078] In step S400 of this embodiment, a dynamic fault threshold surface is established based on a multi-condition benchmark feature library, and equipment operating environment parameter data is introduced to correct the dynamic fault threshold surface in real time, including: Collect equipment operating environment parameter data and construct an environmental parameter dataset, wherein the equipment operating environment parameter data includes ambient temperature, ambient humidity and equipment load rate; Based on the multidimensional space interpolation algorithm, the confidence intervals in the benchmark feature library of each working condition are mapped to the feature-environment parameter space to construct the initial fault threshold surface; An environmental parameter correction factor is introduced, and a weighted calculation is used to dynamically adjust the initial fault threshold surface to obtain a corrected dynamic fault threshold surface. Based on real-time collected equipment operating environment parameters and equipment operating data, the dynamic fault threshold surface is continuously updated through a sliding window mechanism.

[0079] First, it is necessary to collect environmental parameter data of the equipment and build an environmental parameter dataset. This involves collecting ambient temperature, humidity, and equipment load rate during equipment operation, and organizing the data into a structured dataset in chronological order to ensure that the data covers common environmental and load scenarios of the equipment.

[0080] Next, based on multidimensional spatial interpolation algorithms, the confidence intervals in the benchmark feature library for each operating condition need to be mapped to the feature-environment parameter space to construct an initial fault threshold surface. That is, a multidimensional space is constructed using environmental parameters such as temperature, humidity, and load rate, along with feature values, as coordinate axes. Through multidimensional spatial interpolation algorithms such as Kriging interpolation, the confidence intervals of benchmark features for each typical operating condition, such as the normal range of current characteristics under high load and high temperature, are smoothly connected in the space to form an initial fault threshold surface covering the entire scenario. This means that any combination of environment and feature values ​​can correspond to a fault threshold range.

[0081] Next, environmental parameter correction factors need to be introduced, and weighted calculations are used to dynamically adjust the initial fault threshold surface to obtain a corrected dynamic fault threshold surface. Specifically, correction factors for each environmental parameter are determined based on historical data. For example, when environmental humidity > 80%, the impact on insulation-related characteristic thresholds is significant, so a correction factor of 0.9 is set; when environmental temperature fluctuations are small, a correction factor of 1.0 is set. These correction factors are then used to weight the thresholds of the corresponding environmental parameter dimensions in the initial surface. For instance, if the initial threshold upper limit is 100 and the humidity correction factor is 0.9, the initial threshold is adjusted to 90, ultimately obtaining a dynamic fault threshold surface adapted to the environment.

[0082] Furthermore, the dynamic fault threshold surface needs to be continuously updated using a sliding window mechanism based on real-time collected equipment operating environment parameters and equipment operating data. This involves setting a fixed-length sliding window, such as one hour, collecting environmental parameters and equipment operating data within the window in real time, recalculating the feature confidence interval of the local area using the new data, and updating the threshold surface portion within the window coverage area to ensure that the threshold always matches the current equipment operating environment and status.

[0083] Furthermore, it is necessary to combine the equipment health score and the probability distribution of fault types to make multi-dimensional fusion decisions and generate graded fault warning and maintenance strategies.

[0084] In step S400 of this application embodiment, a multi-dimensional fusion decision is made by combining the equipment health score and the fault type probability distribution to generate a graded fault early warning and maintenance strategy, including: By using weighted fusion calculation, the equipment health score and fault type probability distribution are normalized to obtain a comprehensive risk value. The comprehensive risk value is calibrated for operational conditions by referring to the threshold range in the dynamic fault threshold surface that matches the current environmental conditions. Based on the calibrated comprehensive risk value, the early warning levels are divided, including normal level, attention level, early warning level and emergency level; Based on the warning level, the equipment maintenance knowledge base is queried, and corresponding maintenance strategies are generated and maintenance priorities are determined.

[0085] The purpose of the above-mentioned detailed steps in step S400 of the embodiments of this application is to generate accurate early warning levels and maintenance strategies that are in line with actual working conditions by integrating multi-dimensional diagnostic information, so as to achieve accurate transformation from equipment status assessment to operation and maintenance actions, ensure safe equipment operation and reduce maintenance costs.

[0086] To achieve the above objectives, it is first necessary to normalize the equipment health score and fault type probability distribution through weighted fusion calculation to obtain a comprehensive risk value.

[0087] Specifically, the health score needs to be processed in reverse, such as 100 minus the health score, to convert it into a risk baseline value of 0-100; for the probability distribution of failure types, the highest probability value is selected, such as a bearing wear probability of 70%, as the risk contribution value; then, according to preset weights, such as the risk baseline value accounting for 60% and the risk contribution value accounting for 40%, a weighted calculation is performed, and the comprehensive risk value = 0.6 × (100 - health score) + 0.4 × highest failure probability × 100, to obtain the normalized risk value of 0-100.

[0088] Next, the comprehensive risk value needs to be calibrated for operating condition adaptability by referring to the threshold interval in the dynamic fault threshold surface that matches the current environmental conditions. That is, the threshold interval that matches the parameters such as temperature and humidity of the current representative environment and the load rate of the representative operating condition is extracted from the dynamic fault threshold surface. Then, the deviation coefficient between the current threshold interval and the standard operating condition threshold interval is calculated. For example, the deviation coefficient is 1.2 in a high temperature environment. Then, calibrate the comprehensive risk value using the deviation coefficient. The calibrated risk value = comprehensive risk value × deviation coefficient. For example, if the original comprehensive risk value is 60 and the deviation coefficient is 1.2, then the calibrated comprehensive risk value is 72.

[0089] Furthermore, based on the calibrated comprehensive risk value, it is necessary to classify the early warning levels, which include normal, attention, early warning, and emergency levels. Specifically, it is necessary to preset the early warning level classification thresholds, such as normal level (comprehensive risk value ≤ 30), attention level (comprehensive risk value 31-50), early warning level (comprehensive risk value 51-70), and emergency level (comprehensive risk value > 70). The calibrated comprehensive risk value is compared with the early warning level classification thresholds to determine the corresponding level (e.g., a calibrated comprehensive risk value of 72 corresponds to the emergency level). Optionally, the classification can simultaneously refer to abrupt changes in the feature evolution trajectory, such as a sudden increase of 20 points in the comprehensive risk value in a short period of time, which can lead to a leapfrog upgrade of the early warning level.

[0090] Finally, the equipment maintenance knowledge base needs to be queried based on the aforementioned warning levels to generate corresponding maintenance strategies and determine maintenance priorities. Specifically, an equipment maintenance knowledge base needs to be constructed to store standard strategies corresponding to each warning level and fault type, such as "immediately stop and replace the bearing" for the emergency level + bearing wear. Then, based on the current warning level and the highest probability fault type, the maintenance strategy is matched from the equipment maintenance knowledge base according to the warning level. Priority is determined according to "emergency level > warning level > attention level > normal level," with priority given to equipment affecting the critical path of production at the same level; for example, at the same level, the main motor of the production line takes precedence over the auxiliary pump. Following these steps, the graded fault warning and maintenance strategy can be generated.

[0091] Example 2, as Figure 2As shown, based on the same inventive concept as the intelligent equipment fault diagnosis method based on the Modbus protocol provided in Embodiment 1, this embodiment of the invention also provides an intelligent equipment fault diagnosis system based on the Modbus protocol, comprising: The real-time running data acquisition module 11 is used to acquire real-time running data from the target industrial equipment via the Modbus protocol, dynamically adjust the initial sampling frequency based on the equipment's operating status, and perform multiple verifications and compensation corrections on the acquired data to obtain a stable data stream. The feature evolution trajectory construction module 12 is used to perform multi-scale decomposition and feature enhancement processing on the stable data stream, extract the time-frequency domain hybrid feature set, and construct the feature evolution trajectory. The dual-branch diagnostic module 13 is used to input the feature evolution trajectory into the dual-branch diagnostic model that integrates equipment status prediction and fault classification, and output the equipment health score and fault type probability distribution. The multi-dimensional fusion decision module 14 is used to construct a dynamic fault threshold surface based on the equipment's historical operating data and equipment operating environment parameter data, and to perform multi-dimensional fusion decision-making by combining the equipment health score and fault type probability distribution to generate graded fault warning and maintenance strategies.

[0092] Furthermore, the real-time data acquisition module 11 includes the following execution steps: Configure Modbus communication parameters, which include device address, register address, and initial sampling frequency; Monitor the data fluctuation characteristics of the target industrial equipment's current, voltage, and temperature, and adaptively adjust the initial sampling frequency based on the fluctuation amplitude; The collected data is subjected to missing data detection, jump correction and noise filtering, and an abnormal segment is repaired by a sliding window compensation mechanism to form the stable data stream.

[0093] Furthermore, the feature evolution trajectory construction module 12 includes the following execution steps: The stable data stream is subjected to wavelet packet multi-scale decomposition to obtain components of different frequency bands; Extract the time-domain statistical features and frequency-domain energy features of each component to construct an initial feature set; Key features are selected based on grey relational analysis, and feature differentiation is enhanced through feature transformation to generate the time-frequency domain hybrid feature set. The feature evolution trajectory is constructed by combining the time-frequency domain hybrid feature set analysis.

[0094] The feature evolution trajectory is constructed by combining the time-frequency domain hybrid feature set analysis, including: The time-frequency domain hybrid feature set is sorted according to the acquisition time order to obtain a feature sequence, and the feature sequence is segmented using a sliding window of fixed length; Within each sliding window, the mean, standard deviation, and range of the feature values ​​are calculated to form window feature statistics; Based on the dynamic time warping algorithm, the morphological similarity and feature change trend of feature sequences between adjacent sliding windows are analyzed. The peak points, valley points, and inflection points of the feature sequence are extracted to construct a key point sequence of feature evolution, forming the feature evolution trajectory of the equipment state.

[0095] The feature evolution trajectory is constructed by combining the time-frequency domain hybrid feature set analysis, including: The time-frequency domain hybrid feature set is sorted according to the acquisition time order to obtain a feature sequence, and the feature sequence is segmented using a sliding window of fixed length; Within each sliding window, the mean, standard deviation, and range of the feature values ​​are calculated to form window feature statistics; Based on the dynamic time warping algorithm, the morphological similarity and feature change trend of feature sequences between adjacent sliding windows are analyzed. The peak points, valley points, and inflection points of the feature sequence are extracted to construct a key point sequence of feature evolution, forming the feature evolution trajectory of the equipment state.

[0096] Furthermore, the dual-branch diagnostic module 13 includes the following execution steps: The construction steps of the dual-branch diagnostic model include: Based on historical equipment operation data, a set of sample feature evolution trajectories is collected, and the equipment status corresponding to each feature evolution trajectory is labeled to obtain a set of sample equipment statuses. The sample equipment statuses include equipment health status level identifiers and fault type identifiers. Based on machine learning, a network architecture for a dual-branch diagnostic model is constructed, including a device status prediction branch and a device fault classification branch. The dual-branch diagnostic model is trained under supervised supervision using the set of sample feature evolution trajectories and the set of sample device states until the verification accuracy converges, thus completing the model construction.

[0097] Furthermore, the multi-dimensional fusion decision module 14 includes the following execution steps: The steps for constructing the dynamic fault threshold surface include: Based on the historical operating data of the equipment, a set of characteristic evolution trajectories under multiple working conditions is collected, and the equipment status corresponding to each characteristic evolution trajectory is labeled to obtain a sample set of equipment status. Based on the K-means clustering algorithm, the set of feature evolution trajectories is divided into working conditions to determine the typical operating condition categories. For each typical operating condition, a benchmark feature library is constructed, and the confidence interval of the feature values ​​under each condition is calculated. Based on a multi-condition benchmark feature library, a dynamic fault threshold surface is established, and equipment operating environment parameter data is introduced to correct the dynamic fault threshold surface in real time.

[0098] Specifically, based on a multi-condition benchmark feature library, a dynamic fault threshold surface is established, and equipment operating environment parameter data is introduced to correct the dynamic fault threshold surface in real time, including: Collect equipment operating environment parameter data and construct an environmental parameter dataset, wherein the equipment operating environment parameter data includes ambient temperature, ambient humidity and equipment load rate; Based on the multidimensional space interpolation algorithm, the confidence intervals in the benchmark feature library of each working condition are mapped to the feature-environment parameter space to construct the initial fault threshold surface; An environmental parameter correction factor is introduced, and a weighted calculation is used to dynamically adjust the initial fault threshold surface to obtain a corrected dynamic fault threshold surface. Based on real-time collected equipment operating environment parameters and equipment operating data, the dynamic fault threshold surface is continuously updated through a sliding window mechanism.

[0099] Specifically, a multi-dimensional fusion decision is made by combining the equipment health score with the fault type probability distribution to generate a graded fault early warning and maintenance strategy, including: By using weighted fusion calculation, the equipment health score and fault type probability distribution are normalized to obtain a comprehensive risk value. The comprehensive risk value is calibrated for operational conditions by referring to the threshold range in the dynamic fault threshold surface that matches the current environmental conditions. Based on the calibrated comprehensive risk value, the early warning levels are divided, including normal level, attention level, early warning level and emergency level; Based on the warning level, the equipment maintenance knowledge base is queried, and corresponding maintenance strategies are generated and maintenance priorities are determined.

[0100] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0101] Those skilled in the art will understand that embodiments of the present invention can provide methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0102] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0103] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0104] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0105] Although preferred embodiments of the invention have been described, those skilled in the art, once they have learned the basic inventive concept, can make other changes and modifications to these embodiments.

[0106] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for intelligent diagnosis of equipment faults based on the Modbus protocol, characterized in that, The method includes: Real-time operating data is collected from the target industrial equipment using the Modbus protocol. The initial sampling frequency is dynamically adjusted based on the equipment's operating status. The collected data is then subjected to multiple verifications and compensation corrections to obtain a stable data stream. The stable data stream is subjected to multi-scale decomposition and feature enhancement processing to extract a time-frequency domain hybrid feature set and construct a feature evolution trajectory. The feature evolution trajectory is input into a dual-branch diagnostic model that integrates equipment status prediction and fault classification, and the output is the equipment health score and fault type probability distribution. Based on historical equipment operation data and equipment operating environment parameter data, a dynamic fault threshold surface is constructed. The equipment health score and fault type probability distribution are combined to make multi-dimensional fusion decisions and generate graded fault early warning and maintenance strategies.

2. The intelligent equipment fault diagnosis method based on the Modbus protocol according to claim 1, characterized in that, Real-time operational data is collected from the target industrial equipment using the Modbus protocol. The initial sampling frequency is dynamically adjusted based on the equipment's operating status, and the collected data undergoes multiple verifications and compensation corrections to obtain a stable data stream, including: Configure Modbus communication parameters, which include device address, register address, and initial sampling frequency; Monitor the data fluctuation characteristics of the target industrial equipment's current, voltage, and temperature, and adaptively adjust the initial sampling frequency based on the fluctuation amplitude; The collected data is subjected to missing data detection, jump correction and noise filtering, and an abnormal segment is repaired by a sliding window compensation mechanism to form the stable data stream.

3. The intelligent equipment fault diagnosis method based on the Modbus protocol according to claim 1, characterized in that, The stable data stream is subjected to multi-scale decomposition and feature enhancement processing to extract a time-frequency domain hybrid feature set and construct a feature evolution trajectory, including: The stable data stream is subjected to wavelet packet multi-scale decomposition to obtain components of different frequency bands; Extract the time-domain statistical features and frequency-domain energy features of each component to construct an initial feature set; Key features are selected based on grey relational analysis, and feature differentiation is enhanced through feature transformation to generate the time-frequency domain hybrid feature set. The feature evolution trajectory is constructed by combining the time-frequency domain hybrid feature set analysis.

4. The intelligent equipment fault diagnosis method based on the Modbus protocol according to claim 3, characterized in that, The feature evolution trajectory is constructed by combining the time-frequency domain hybrid feature set analysis, including: The time-frequency domain hybrid feature set is sorted according to the acquisition time order to obtain a feature sequence, and the feature sequence is segmented using a sliding window of fixed length; Within each sliding window, the mean, standard deviation, and range of the feature values ​​are calculated to form window feature statistics; Based on the dynamic time warping algorithm, the morphological similarity and feature change trend of feature sequences between adjacent sliding windows are analyzed. The peak points, valley points, and inflection points of the feature sequence are extracted to construct a key point sequence of feature evolution, forming the feature evolution trajectory of the equipment state.

5. The intelligent equipment fault diagnosis method based on the Modbus protocol according to claim 1, characterized in that, The steps for constructing the dual-branch diagnostic model include: Based on historical equipment operation data, a set of sample feature evolution trajectories is collected, and the equipment status corresponding to each feature evolution trajectory is labeled to obtain a set of sample equipment statuses. The sample equipment statuses include equipment health status level identifiers and fault type identifiers. Based on machine learning, a network architecture for a dual-branch diagnostic model is constructed, including a device status prediction branch and a device fault classification branch. The dual-branch diagnostic model is trained under supervision using the set of sample feature evolution trajectories and the set of sample device states until the verification accuracy converges, thus completing the model construction.

6. The intelligent equipment fault diagnosis method based on the Modbus protocol according to claim 1, characterized in that, The steps for constructing the dynamic fault threshold surface include: Based on the historical operating data of the equipment, a set of characteristic evolution trajectories under multiple working conditions is collected, and the equipment status corresponding to each characteristic evolution trajectory is labeled to obtain a sample set of equipment status. Based on the K-means clustering algorithm, the set of feature evolution trajectories is divided into working conditions to determine the typical operating condition categories. For each typical operating condition, a benchmark feature library is constructed, and the confidence interval of the feature values ​​under each condition is calculated. Based on a multi-condition benchmark feature library, a dynamic fault threshold surface is established, and equipment operating environment parameter data is introduced to correct the dynamic fault threshold surface in real time.

7. The intelligent equipment fault diagnosis method based on the Modbus protocol according to claim 6, characterized in that, Based on a multi-condition benchmark feature library, a dynamic fault threshold surface is established, and equipment operating environment parameter data is introduced to perform real-time correction on the dynamic fault threshold surface, including: Collect equipment operating environment parameter data and construct an environmental parameter dataset, wherein the equipment operating environment parameter data includes ambient temperature, ambient humidity and equipment load rate; Based on the multidimensional space interpolation algorithm, the confidence intervals in the benchmark feature library of each working condition are mapped to the feature-environment parameter space to construct the initial fault threshold surface; An environmental parameter correction factor is introduced, and a weighted calculation is used to dynamically adjust the initial fault threshold surface to obtain a corrected dynamic fault threshold surface. Based on real-time collected equipment operating environment parameters and equipment operating data, the dynamic fault threshold surface is continuously updated through a sliding window mechanism.

8. The intelligent equipment fault diagnosis method based on the Modbus protocol according to claim 1, characterized in that, By combining the equipment health score with the fault type probability distribution, a multi-dimensional fusion decision is made to generate a graded fault early warning and maintenance strategy, including: By using weighted fusion calculation, the equipment health score and fault type probability distribution are normalized to obtain a comprehensive risk value. The comprehensive risk value is calibrated for operational conditions by referring to the threshold range in the dynamic fault threshold surface that matches the current environmental conditions. Based on the calibrated comprehensive risk value, the early warning levels are divided, including normal level, attention level, early warning level and emergency level; Based on the warning level, the equipment maintenance knowledge base is queried, and corresponding maintenance strategies are generated and maintenance priorities are determined.

9. A device fault intelligent diagnosis system based on the Modbus protocol, characterized in that, The system is used to implement the intelligent equipment fault diagnosis method based on the Modbus protocol as described in any one of claims 1-8, including: The real-time running data acquisition module is used to acquire real-time running data from the target industrial equipment via the Modbus protocol, dynamically adjust the initial sampling frequency based on the equipment's operating status, and perform multiple verifications and compensation corrections on the acquired data to obtain a stable data stream. The feature evolution trajectory construction module is used to perform multi-scale decomposition and feature enhancement processing on the stable data stream, extract the time-frequency domain hybrid feature set, and construct the feature evolution trajectory. The dual-branch diagnostic module is used to input the feature evolution trajectory into the dual-branch diagnostic model that integrates equipment status prediction and fault classification, and output the equipment health score and fault type probability distribution. The multi-dimensional fusion decision module is used to construct a dynamic fault threshold surface based on the equipment's historical operating data and equipment operating environment parameter data. It combines the equipment health score and fault type probability distribution to perform multi-dimensional fusion decision-making and generate graded fault early warning and maintenance strategies.

Images