Rolling bearing fault diagnosis method based on SDP image

Abstract

The invention relates to the technical field of rolling bearing diagnosis, and discloses a rolling bearing fault diagnosis method based on an SDP image, and the method comprises the following steps: S1, a multi-source data collaborative collection stage; s2, an SDP image generation and enhancement stage; s3, a dual-path feature extraction and model training stage; s4, a real-time diagnosis and closed-loop optimization stage; s5, a fault traceability and report generation stage; the objective of the invention is to solve the problems that an existing fault diagnosis method mostly depends on single-type operation data, complex features of bearing faults are difficult to comprehensively capture, and missed judgment of slight faults or composite faults is easily caused by data dimension limitation; meanwhile, a traditional method mostly adopts a single feature extraction mode, either only simple features of a shallow layer can be excavated, or the stability of basic features is ignored due to excessive dependence on deep features, the identification requirements of different fault types cannot be considered, and the adaptability to operation condition changes is poor. And fault type misjudgment or inaccurate severity evaluation is easy to occur.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of rolling bearing diagnosis, and in particular to a rolling bearing fault diagnosis method based on an SDP image. BACKGROUND

[0002] The rolling bearing fault diagnosis of the SDP image refers to the following steps: collecting multi-dimensional running data of a rolling bearing during operation, converting a one-dimensional mechanical signal into a two-dimensional visual image by using a symmetric point mode (SDP) technology, analyzing features such as a texture and a gray scale distribution related to a fault in the image, identifying whether a fault exists in an inner ring, an outer ring, a rolling element or the like of the bearing and a severity of the fault, and essentially converting a mechanical fault feature into an image feature for accurate identification.

[0003] Existing fault diagnosis methods mostly rely on a single type of running data, and it is difficult to comprehensively capture complex features of a bearing fault, and a slight fault or a composite fault may be missed due to a limited data dimension. Meanwhile, a single feature extraction method is mostly used in traditional methods, and either only shallow simple features can be excavated or a deep feature is excessively relied on and a stability of basic features is ignored, so that different fault types cannot be recognized, adaptability to a running condition change is poor, and a fault type misjudgment or an inaccurate severity evaluation problem is likely to occur. SUMMARY

[0004] The present application provides a rolling bearing fault diagnosis method based on an SDP image, so as to solve the problem that existing fault diagnosis methods mostly rely on a single type of running data, and it is difficult to comprehensively capture complex features of a bearing fault, a slight fault or a composite fault may be missed due to a limited data dimension, and a single feature extraction method is mostly used in traditional methods, and either only shallow simple features can be excavated or a deep feature is excessively relied on and a stability of basic features is ignored, so that different fault types cannot be recognized, adaptability to a running condition change is poor, and a fault type misjudgment or an inaccurate severity evaluation problem is likely to occur.

[0005] In order to achieve the above-mentioned purpose, the present application adopts the following technical scheme:

[0006] The rolling bearing fault diagnosis method based on the SDP image comprises the following steps:

[0007] S1, a multi-source data cooperative collection stage, multi-dimensional sensor groups are used to collect bearing vibration, temperature, rotating speed and running noise data, the data are stored and labeled with relevant attributes by time stamp alignment, abnormal data are removed and missing data are completed;

[0008] S2, SDP image generation and enhancement stage, filtering and standardization preprocessing of collected data, combination of multi-source parameters to dynamically adjust SDP generation parameters to generate images, optimization of image quality through histogram equalization and Gaussian blur;

[0009] S3, dual-path feature extraction and model training stage, extraction of manual features and deep features of SDP images, construction of two-level diagnosis model and completion of training with cross-entropy loss function;

[0010] S4, real-time diagnosis and closed-loop optimization stage, input of real-time processed SDP images into the model to output diagnosis results, monthly archiving of data and iterative optimization of model parameters through gradient descent method, triggering of graded early warning according to threshold;

[0011] S5, fault tracing and report generation stage, correlation of multi-source data time series to locate fault occurrence period and analyze causes, integration of various data to generate standardized diagnosis report.

[0012] As a further improved scheme of the technical solution: the multi-source data cooperative collection stage includes: deploying a multi-dimensional sensor group, the multi-dimensional sensor group includes a three-axis acceleration sensor, a temperature sensor, an incremental optical encoder, and a microphone array, which respectively collect bearing X / Y / Z three-axis vibration signals, outer ring temperature, real-time speed, and running noise, the vibration sensor sampling frequency is 20kHz, the temperature sensor measurement range is-20℃ to 150℃, the speed sensor accuracy is ±1rpm, and the acoustic sensor frequency range is 20Hz to 20kHz; the multi-source data is aligned by time stamp through a data synchronous storage unit, stored in 10 minutes as a data segment, and labeled with bearing model, running condition, and cumulative running time; the abnormal data preliminary screening tool adopts a 3σ criterion to remove abnormal points in the vibration signal with a jump greater than 2 times the rated vibration value, and a linear interpolation method is used to complete the missing data.

[0013] As a further improved scheme of the technical solution: the SDP image generation and enhancement stage includes: performing first-order low-pass filtering on the vibration signal through a signal preprocessing tool, with a filter cutoff frequency of 500Hz, and performing standardization processing on the temperature and speed data; generating an SDP image through a dynamic parameter SDP generation unit, and the core formula is: , wherein is the gray value of the SDP image coordinates x and y, is the normalized vibration signal amplitude, is the signal sampling point index, is the total number of signal sampling points of a single data segment, , The coordinate value of the SDP point is a coordinate value of the SDP point, the point distance d is in a range of 0.5mm to 2mm, the angle step theta is in a range of 5° to 15°, and sigma is a smoothing coefficient and is in a range of 0.1 to 0.5; the generated SDP image is subjected to histogram equalization processing by the image enhancement unit, and Gaussian blur processing is performed by using a Gaussian kernel with a standard deviation of 0.8.

[0014] As a further improved scheme of the technical solution, the double-path feature extraction and model training stage includes: extracting features through a double-path feature extraction unit, a hand-crafted feature extraction tool extracts the gray level co-occurrence matrix features and seven Hu invariant moment features of the SDP image, and a deep feature extraction network adopts a CNN+Transformer hybrid architecture, the CNN part includes three convolution layers and two pooling layers, and the Transformer layer includes four attention heads; a two-level diagnosis model is constructed, the first level is a normal / abnormal binary classification model, and the second level is an inner ring fault / outer ring fault / rolling element fault multi-classification model, and the two-level models share the deep feature extraction network; the model is trained by using a cross-entropy loss function, and the formula is: wherein is a loss function value, is the total number of training samples, is the number of fault categories, when =2 for binary classification, =4 for multi-classification, is the true label of the ith sample, is the probability that the ith sample is predicted as the cth class.

[0015] As a further improved scheme of the technical solution, the real-time diagnosis and closed-loop optimization stage includes: the trained two-level diagnosis model is carried by an online diagnosis reasoning unit, the real-time processed SDP image is received and the fault type, fault confidence and fault severity score are output, the fault confidence threshold is greater than or equal to 0.9 for effective diagnosis, and the fault severity score ranges from 0 to 100 points; the model iteration optimization unit archives the full amount of diagnosis data and fault feedback results monthly, takes the diagnosis error as the optimization target, adjusts the model parameters by the gradient descent method, and the iteration formula is: wherein is the model parameter set after iteration, is the current model parameter set before iteration, is a learning rate and is 0.001, is the gradient of the loss function at the current parameters; the hierarchical early warning triggering unit triggers a warning, a first-level warning is triggered when the fault confidence is between 0.7 and 0.9, and a second-level warning is triggered when the fault confidence is greater than or equal to 0.9 and the severity score is less than 60 points.

[0016] As a further improvement of the technical solution: the fault tracing and report generation stage includes: through the time sequence data tracing unit, the time sequence change track of the multi-source data is associated, the fault initial occurrence period is located, the evolution process of the characteristic texture in the SDP image is compared, and the fault cause is analyzed; through the standardized report generation unit, the bearing basic information, the operation condition statistics, the SDP image feature comparison chart, the fault type, the fault severity, the maintenance suggestion and the historical fault comparison data are integrated, and the printable standardized diagnosis report is generated.

[0017] Compared with the prior art, the beneficial effects of the present application are:

[0018] 1、The present application combines multi-source data collaborative collection with SDP image dynamic generation to convert one-dimensional running signals into visual two-dimensional images, which can more intuitively and comprehensively highlight the exclusive features of various bearing faults, avoiding the limitations of single data dimension; at the same time, the dual-path feature extraction and two-level diagnosis model architecture design not only retain the stable recognition ability of traditional features, but also can mine deep complex fault features, effectively improving the accuracy of fault recognition, solving the problem of insufficient recognition of slight faults and compound faults in traditional diagnosis methods, and ensuring accurate determination of fault type and severity.

[0019] 2、The real-time diagnosis and closed-loop optimization mechanism provided by the present application can quickly output diagnosis results and trigger graded early warning, providing timely reference for on-site operation and maintenance, and shortening the fault disposal period; in the long term, the model can be automatically iteratively optimized through historical data, dynamically adapting to changes in bearing operation state, reducing manual maintenance cost; in addition, the standardized report and fault tracing function provide systematic data support for operation and maintenance decision-making, help to analyze the causes of faults, and form a standardized operation and maintenance process, which not only ensures the safety and continuity of equipment operation, but also further reduces the overall cost of operation and maintenance.

[0020] The above description is only a summary of the technical solution of the present application, in order to more clearly understand the technical means of the present application, and the content of the specification can be implemented, the following is a preferred embodiment of the present application, and the detailed description is as follows. The specific embodiments of the present application are given in detail by the following examples and their drawings. BRIEF DESCRIPTION OF DRAWINGS

[0021] The drawings described herein are used to provide further understanding of the present application, and form a part of the present application. The schematic embodiments of the present application and their descriptions are used to explain the present application, and do not constitute an improper limitation on the present application. In the drawings:

[0022] Figure 1 It is a method structure diagram of the rolling bearing fault diagnosis method based on SDP image. DETAILED DESCRIPTION

[0023] The principles and characteristics of the present application are described below in conjunction with the drawings, and the examples are used only to explain the present application and are not intended to limit the scope of the present application. In the following paragraphs, the present application is described in more detail by way of example with reference to the accompanying drawings. It should be noted that the drawings are all very simplified and use non-precise proportions, only for the purpose of facilitating, clarifying and assisting in the description of the embodiments of the present application.

[0024] Referring to Figure 1 In the embodiments of the present application, the rolling bearing fault diagnosis method based on SDP image comprises the following steps:

[0025] S1, multi-source data collaborative acquisition stage, the bearing vibration, temperature, speed, running noise data are collected by multi-dimensional sensor group, and the relevant attributes are stored and labeled by time stamp alignment, the abnormal data are eliminated and the missing data are completed;

[0026] S2, SDP image generation and enhancement stage, the collected data are filtered and standardized pretreated, the SDP generation parameters are dynamically adjusted combined with multi-source parameters to generate images, and the image quality is optimized by histogram equalization and Gaussian blur;

[0027] S3, double-path feature extraction and model training stage, the manual features and deep features of the SDP image are extracted, the two-level diagnosis model is constructed, and the training is completed by using cross-entropy loss function;

[0028] S4, real-time diagnosis and closed-loop optimization stage, the SDP image after real-time processing is input into the model to output the diagnosis result, the data are archived monthly and the model parameters are iteratively optimized by gradient descent method, and the graded early warning is triggered according to the threshold value;

[0029] S5, fault tracing and report generation stage, the multi-source data time sequence change is associated to locate the fault occurrence period and analyze the causes, and various data are integrated to generate standardized diagnosis report;

[0030] Specifically, the core role of each stage is:

[0031] S1, multi-source data collaborative acquisition stage: obtaining multi-dimensional original data of bearing operation provides basic data support for subsequent diagnosis;

[0032] S2, SDP image generation and enhancement stage: converting one-dimensional signal into two-dimensional visual image, optimizing image quality to highlight fault features;

[0033] S3, double-path feature extraction and model training stage: mining fault features in the image, constructing and training the model capable of accurately identifying faults;

[0034] S4, real-time diagnosis and closed-loop optimization stage: realizing real-time determination of faults, improving long-term diagnosis accuracy of the model through iterative optimization;

[0035] S5, fault tracing and report generation stage: trace the root cause of the fault, and generate a standardized report to provide decision basis for operation and maintenance.

[0036] The multi-source data collaborative collection stage includes: deploying a multi-dimensional sensor group, the multi-dimensional sensor group including a three-axis acceleration sensor, a temperature sensor, an incremental photoelectric encoder, and a microphone array, respectively collecting bearing X / Y / Z three-axis vibration signals, outer ring temperature, real-time rotating speed, and running noise, the vibration sensor sampling frequency being 20 kHz, the temperature sensor measurement range being -20℃ to 150℃, the rotating speed sensor accuracy being ±1 rpm, and the acoustic sensor frequency range being 20 Hz to 20 kHz; the multi-source data is aligned by time stamp through a data synchronous storage unit, and stored in 10 minutes as a data segment, and the bearing model, running condition, and cumulative running time are labeled; the abnormal data preliminary screening tool adopts a 3σ criterion to eliminate abnormal points in the vibration signal with a jump greater than 2 times the rated vibration value, and adopts a linear interpolation method to complete the missing data;

[0037] Specifically, the multi-dimensional sensor group includes a three-axis acceleration sensor, a temperature sensor, an incremental photoelectric encoder, and a microphone array; the three-axis acceleration sensor collects X, Y, and Z three-axis vibration signals (sampling frequency 20 kHz) to capture fault features in bearing mechanical vibration; the temperature sensor measures the outer ring temperature (range -20℃ to 150℃) to reflect the bearing friction loss state; the incremental photoelectric encoder obtains real-time rotating speed (accuracy ±1 rpm) to provide basis for condition adaptation; and the microphone array collects running noise (frequency range 20 Hz to 20 kHz) to assist in verifying the fault features captured by the vibration signal;

[0038] The data synchronous storage unit aligns the multi-source data by time stamp (10 minutes as a data segment), labels the bearing model, running condition, and cumulative running time, and ensures data time sequence consistency and traceability;

[0039] The abnormal data preliminary screening tool adopts a 3σ criterion to eliminate abnormal points in the vibration signal with a jump greater than 2 times the rated vibration value, and adopts a linear interpolation method to complete the missing data, to preliminarily purify data quality.

[0040] The SDP image generation and enhancement stage includes: performing first-order low-pass filtering on the vibration signal through a signal preprocessing tool, the filtering cutoff frequency being 500 Hz, and performing standardization processing on the temperature and rotating speed data; generating an SDP image through a dynamic parameter SDP generation unit, the core formula being: wherein is the gray value of the SDP image coordinates x and y, is the normalized vibration signal amplitude, is the signal sampling point index, The total number of signal sampling points in a single data segment. , The coordinates of the SDP points are given, the point spacing d ranges from 0.5mm to 2mm, the angle step θ ranges from 5° to 15°, and σ is the smoothing coefficient, ranging from 0.1 to 0.5. The generated SDP image is processed by histogram equalization through the image enhancement unit, and Gaussian blurring is performed using a Gaussian kernel with a standard deviation of 0.8.

[0041] Specifically, the signal preprocessing tools perform first-order low-pass filtering (cutoff frequency 500Hz) on the vibration signal to filter high-frequency noise, and standardize the temperature and rotation speed data to eliminate dimensional differences, providing clean input for SDP image generation.

[0042] Dynamic parameter SDP generation unit: converts one-dimensional vibration signals into two-dimensional SDP images, and dynamically adjusts the generated parameters through multi-source parameters to ensure accurate visualization of fault characteristics;

[0043] Core formula: Formula annotation: SDP image coordinates The grayscale value at the location (range 0-255) reflects the strength distribution of fault characteristics based on the difference in grayscale value. This represents the normalized vibration signal amplitude (mapped to the 0-1 interval). For signal sampling point index, The total number of signal sampling points in a single data segment; , The x and y coordinates of the SDP points are given. The point spacing d (0.5 mm to 2 mm) and angle step θ (5° to 15°) are dynamically adjusted with the rotational speed (the higher the rotational speed, the smaller d and θ are); σ is the smoothing coefficient (0.1 to 0.5), which is dynamically adjusted with the temperature (the higher the temperature, the smaller σ is); The formula's function is to break through the limitations of fixed parameters in traditional SDP images, and to adjust the generated parameters through multi-source data linkage, so that the image can adapt to different operating conditions and accurately highlight the unique texture features of inner ring, outer ring, and rolling element faults;

[0044] Image enhancement unit: Optimizes image contrast through histogram equalization and performs Gaussian blurring using a Gaussian kernel with a standard deviation of 0.8 to suppress noise while preserving core fault features and improve image quality.

[0045] The double-path feature extraction and model training stage comprises: extracting features by a double-path feature extraction unit, a hand-crafted feature extraction tool extracts a gray level co-occurrence matrix feature and seven Hu invariant moment features of the SDP image, a deep feature extraction network adopts a CNN+Transformer hybrid architecture, the CNN part comprises three convolutional layers and two pooling layers, and the Transformer layer comprises four attention heads; a two-level diagnosis model is constructed, the first level is a normal / abnormal binary classification model, and the second level is an inner ring fault, outer ring fault and rolling element fault multi-classification model, and the two-level models share the deep feature extraction network; a cross-entropy loss function is used to train the model, and the formula is: , wherein is a loss function value, is a total number of training samples, is a number of fault categories, when binary classification = 2, and when multi-classification = 4, is a real label of the i-th sample, is a probability that the i-th sample is predicted to be the c-th category;

[0046] Specifically, the double-path feature extraction unit adopts a hand-crafted feature and deep feature double-path parallel architecture, and both stability and generalization ability are considered;

[0047] The hand-crafted feature extraction tool extracts a gray level co-occurrence matrix feature (six features such as contrast, correlation and energy) and seven Hu invariant moment features of the SDP image, reflects image texture distribution and shape characteristics, and is suitable for simple fault rapid identification;

[0048] The deep feature extraction network adopts a CNN and Transformer hybrid architecture, the CNN part comprises three convolutional layers and two pooling layers, and extracts local fault texture features (such as crack lines and wear spots); the Transformer layer comprises four attention heads, captures global association between local features through a self-attention mechanism, and solves the problem that a single CNN is difficult to capture cross-regional fault features;

[0049] The two-level diagnosis model comprises two-level progressive models of binary classification and multi-classification, the first level model judges bearing normality and abnormality, the second level model distinguishes inner ring fault, outer ring fault and rolling element fault, and the two-level models share the deep feature extraction network to reduce calculation overhead;

[0050] The cross-entropy loss function is: , and the formula is marked as: is a loss function value (reflecting deviation of a predicted result from a real label, and the smaller the value, the better the model performance); is a total number of training samples; C is a number of fault categories (C=2 when binary classification, and C=4 when multi-classification); Yic is the true label of the ith sample (one-hot encoding, 1 if it belongs to the cth class, otherwise 0); Pici is the probability of the ith sample being predicted as the cth class (0-1 interval); Formula function: as the optimization objective of model training, guide the model to learn the mapping relationship between fault features and categories, and improve the classification accuracy.

[0051] The real-time diagnosis and closed-loop optimization stage includes: through the online diagnosis reasoning unit, the two-level diagnosis model trained is carried, the SDP image after real-time processing is received, and the fault type, fault confidence, and fault severity score are outputted, the fault confidence threshold is greater than or equal to 0.9 for effective diagnosis, and the fault severity score ranges from 0 to 100 points; through the model iterative optimization unit, the full amount of diagnosis data and fault feedback results are archived monthly, the diagnosis error is taken as the optimization objective, the model parameters are adjusted through the gradient descent method, and the iterative formula is: , wherein is the model parameter set after iteration, is the current model parameter set before iteration, is the learning rate, which is 0.001, is the gradient of the loss function at the current parameter; through the hierarchical early warning triggering unit, when the fault confidence is between 0.7 and 0.9, the first level early warning is triggered, and when the fault confidence is greater than or equal to 0.9 and the severity score is less than 60 points, the second level early warning is triggered;

[0052] Specifically, the online diagnosis reasoning unit: carries the two-level diagnosis model trained, receives the SDP image after real-time processing, outputs the fault type, fault confidence (threshold greater than or equal to 0.9 for effective diagnosis), and fault severity score (0-100 points, the lower the score, the more serious the fault), and provides instant reference for field operation and maintenance;

[0053] The model iterative optimization unit has self-evolution ability and improves the model precision through continuous iteration; the iterative formula is: , formula annotation: is the model parameter set after iteration (including CNN convolution kernel weight, Transformer attention head parameter, etc.); is the current model parameter set before iteration; is the learning rate (value 0.001), which controls the parameter adjustment step; The gradient of the loss function at the current parameters reflects the influence direction of parameter changes on the loss function; the formula acts: archive full-quantity diagnostic data and fault feedback results by month, take diagnostic error as the optimization target, dynamically adjust the model parameters, adapt to the changes of fault features in the bearing aging process, and long-term maintain high diagnostic precision; the hierarchical early warning triggering unit: when the fault confidence is between 0.7 and 0.9, a first-level early warning (prompting close monitoring) is triggered, and when the confidence is greater than or equal to 0.9 and the severity score is less than 60 points, a second-level early warning (suggestion of shutdown maintenance) is triggered, realizing the hierarchical control of fault risk.

[0054] The fault tracing and report generation stage includes: associating the time sequence change track of multi-source data through the time sequence data tracing unit, locating the initial fault occurrence period, corresponding to the evolution process of the feature texture in the SDP image, and analyzing the fault causes; through the standardized report generation unit, integrating bearing basic information, running condition statistics, SDP image feature comparison chart, fault type, fault severity, maintenance suggestion, historical fault comparison data, and generating a printable standardized diagnostic report;

[0055] Specifically, the time sequence data tracing unit: associates the time sequence change track of multi-source data, locates the initial fault occurrence period (such as the vibration amplitude mutation time point, the temperature rising inflection point), corresponds to the evolution process of the feature texture in the SDP image, and analyzes the fault causes (such as overload operation, insufficient lubrication, installation deviation);

[0056] The standardized report generation unit: integrates bearing basic information (model, cumulative running time), running condition statistics (average load, average speed, temperature distribution), SDP image feature comparison chart (normal vs. fault), fault type and severity, maintenance suggestion, historical fault comparison data, and generates a printable standardized diagnostic report, providing a system reference for operation and maintenance decision-making.

[0057] The use method and working principle of the application are:

[0058] Use method: when using, the sensor group, data storage and processing related equipment need to be deployed first, the system initialization and parameter configuration are completed, and it is ensured that each module can work normally; then start the data acquisition function, let the system automatically acquire multi-dimensional data in the bearing running process and pre-process, then the system will automatically convert the processed data into SDP images and optimize the quality, then complete fault recognition through the preset feature extraction path and diagnostic model, users can real-time view diagnostic results and early warning information, finally carry out operation and maintenance disposal according to the generated standardized report, the system will automatically archive data and complete the iteration optimization of the model, without manual intervention in the core diagnostic process.

[0059] Working principle: Based on multi-source data collaborative collection, firstly, one-dimensional signal of bearing operation is converted into two-dimensional SDP image rich in fault characteristics, and the identification degree of fault characteristics is improved through image enhancement technology; secondly, a double-path feature extraction method is adopted to fully mine shallow manual features and deep complex features in the image, and a two-level diagnosis model is used to realize accurate classification and judgment of faults; at the same time, through a closed-loop iteration mechanism, historical data are used to continuously optimize model parameters, adapt to the dynamic changes of bearing operation state, and timely feedback fault risk combined with a hierarchical early warning mechanism; finally, through time series data correlation analysis, the fault root cause is traced back and a report is generated, forming a complete logical chain of data collection, image conversion, feature extraction, fault diagnosis, and optimization feedback, and realizing efficient and accurate diagnosis of bearing faults.

[0060] The above is only a preferred embodiment of the present application, and does not limit the present application in any form; any person skilled in the art can easily implement the present application according to the description and the above; however, any equivalent changes, modifications and evolutions made by those skilled in the art within the scope of the technical solutions of the present application, using the above disclosed technical content, are equivalent embodiments of the present application; at the same time, any equivalent changes, modifications and evolutions made according to the essence of the present application to the above embodiments are still within the protection scope of the technical solutions of the present application.

Claims

1. A method for diagnosing rolling bearing faults based on SDP images, characterized in that, Includes the following steps: S1. Multi-source data collaborative acquisition stage: bearing vibration, temperature, speed and operating noise data are collected through multi-dimensional sensor groups, stored with timestamp alignment and relevant attributes are labeled, abnormal data are removed and missing data is filled in; In the S2 and SDP image generation and enhancement stages, the collected data is filtered and standardized preprocessed, and the SDP generation parameters are dynamically adjusted in combination with multi-source parameters to generate images. The image quality is optimized through histogram equalization and Gaussian blur. S3, Dual-path feature extraction and model training stage: Extract handcrafted features and deep features from SDP images, construct a two-level diagnostic model and complete training using the cross-entropy loss function; S4. Real-time diagnosis and closed-loop optimization stage: Input the real-time processed SDP image into the model to output the diagnosis result, archive the data monthly and iteratively optimize the model parameters through gradient descent method, and trigger graded early warning according to the threshold. S5. In the fault tracing and report generation stage, the fault occurrence time is located by correlating the time-series changes of multi-source data and analyzing the causes. Various types of data are integrated to generate a standardized diagnostic report.

2. The rolling bearing fault diagnosis method based on SDP images according to claim 1, characterized in that, The multi-source data collaborative acquisition stage includes: deploying a multi-dimensional sensor group, which includes a triaxial accelerometer, a temperature sensor, an incremental photoelectric encoder, and a microphone array, to collect bearing X / Y / Z triaxial vibration signals, outer ring temperature, real-time rotational speed, and operating noise, respectively. The vibration sensor sampling frequency is 20kHz, the temperature sensor measurement range is -20℃ to 150℃, the rotational speed sensor accuracy is ±1rpm, and the acoustic sensor frequency range is 20Hz to 20kHz; aligning the multi-source data according to timestamps through a data synchronization storage unit, storing data in 10-minute segments, and labeling the bearing model, operating condition, and cumulative running time; using an anomaly data screening tool, the 3σ criterion is used to remove anomalies in the vibration signal with jumps greater than twice the rated vibration value, and linear interpolation is used to complete the missing data.

3. The rolling bearing fault diagnosis method based on SDP images according to claim 1, characterized in that, The SDP image generation and enhancement stage includes: performing first-order low-pass filtering on the vibration signal using a signal preprocessing tool, with a filter cutoff frequency of 500Hz; standardizing the temperature and rotation speed data; and generating an SDP image using a dynamic parameter SDP generation unit, with the core formula being: ,in The gray values ​​at coordinates x and y in the SDP image. The normalized vibration signal amplitude, For signal sampling point index, The total number of signal sampling points in a single data segment. , The coordinates of the SDP points are given, the point spacing d ranges from 0.5 mm to 2 mm, the angle step θ ranges from 5° to 15°, and σ is the smoothing coefficient, ranging from 0.1 to 0.

5. The generated SDP image is processed by histogram equalization through the image enhancement unit, and Gaussian blurring is performed using a Gaussian kernel with a standard deviation of 0.

8.

4. The rolling bearing fault diagnosis method based on SDP images according to claim 1, characterized in that, The dual-path feature extraction and model training phase includes: extracting features through a dual-path feature extraction unit; extracting gray-level co-occurrence matrix features and 7 Hu invariant moment features from the SDP image using a manual feature extraction tool; employing a CNN+Transformer hybrid architecture for the deep feature extraction network, with the CNN containing 3 convolutional layers and 2 pooling layers, and the Transformer layer containing 4 attention heads; constructing a two-level diagnostic model: a first-level binary classification model for normal and abnormal cases, and a second-level multi-class classification model for inner ring faults, outer ring faults, and rolling element faults, with both levels sharing the deep feature extraction network; and training the model using the cross-entropy loss function, the formula of which is: ,in The value of the loss function. The total number of training samples, The number of fault categories, in binary classification. =2, multi-class classification =4, Let i be the true label of the i-th sample. Let be the probability that the i-th sample is predicted to be of class c.

5. The rolling bearing fault diagnosis method based on SDP images according to claim 1, characterized in that, The real-time diagnosis and closed-loop optimization phase includes: using an online diagnostic inference unit equipped with a trained two-level diagnostic model, receiving real-time processed SDP images and outputting fault type, fault confidence, and fault severity score. A fault confidence threshold ≥0.9 is considered a valid diagnosis, and the fault severity score ranges from 0 to 100. The model iteration optimization unit archives all diagnostic data and fault feedback results monthly, using diagnostic error as the optimization objective, and adjusts model parameters using gradient descent. The iteration formula is as follows: ,in The parameter set of the iterated model. The current model parameter set before iteration. The learning rate is 0.

001. The gradient of the loss function at the current parameters is given. Warnings are issued through a tiered warning triggering unit. A first-level warning is triggered when the fault confidence is between 0.7 and 0.9, and a second-level warning is triggered when the fault confidence is ≥0.9 and the severity score is <60.

6. The rolling bearing fault diagnosis method based on SDP images according to claim 1, characterized in that, The fault tracing and report generation stage includes: using the time-series data tracing unit to associate the time-series change trajectory of multi-source data, locating the initial occurrence time of the fault, corresponding to the evolution process of feature textures in the SDP image, and analyzing the fault causes; and using the standardized report generation unit to integrate basic bearing information, operating condition statistics, SDP image feature comparison map, fault type, fault severity, maintenance suggestions, and historical fault comparison data to generate a printable standardized diagnostic report.

Images