Equipment evaluation and maintenance suggestion method and system based on multi-modal data fusion

Abstract

The invention provides an equipment evaluation and maintenance suggestion method and system based on multi-modal data fusion, and relates to the technical field of equipment evaluation.According to the method, fault feature retrieval and maintenance suggestion intelligent output are achieved by constructing a classification database containing multi-modal historical fault information, combining weight factors and utilizing a large language model; a vibration signal and a temperature field are effectively fused, a bearing, a strain gauge, a motor, a guide rail and a lead screw are analyzed one by one, feature vectors are extracted, correlation analysis is performed on the feature vectors, multi-source features are fused, an equipment fault evaluation index is generated, and a numerical control machine tool state and maintenance suggestion is output based on the equipment fault evaluation index. The system improves the accuracy and sensitivity of fault diagnosis, can dynamically reflect the health state of equipment, assists a worker in taking maintenance measures in time, and remarkably improves the operation safety and production efficiency of the equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment evaluation, and in particular to a method and system for equipment evaluation and maintenance suggestion based on multimodal data fusion. Background Art

[0002] With the continuous improvement of automation and intelligence levels for CNC machine tools in modern manufacturing, equipment operating status monitoring and fault diagnosis have become critical links in ensuring production efficiency and processing quality. Traditional fault diagnosis methods often rely on single sensor data or empirical judgments, which cannot fully reflect the complex operating status of machine tools. This is especially true when multiple fault types coexist or the fault signals are weak, which limits the accuracy and timeliness of diagnosis. Furthermore, CNC machine tool fault information is diverse, covering multimodal data such as vibration signals and temperature fields. How to efficiently integrate and utilize this heterogeneous data makes it difficult to extract effective features. Data or empirical judgments based on a single sensor are less accurate, making it difficult to achieve scientific and efficient maintenance decisions.

[0003] The above information disclosed in the Background section is for understanding the background of the disclosure, and may also include prior art information that is known to a person of ordinary skill in the art. Summary of the Invention

[0004] The purpose of the present invention is to provide a multi-modal data fusion equipment evaluation and maintenance suggestion method and system to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A multimodal data fusion equipment evaluation and maintenance recommendation method, comprising the following steps:

[0007] S1. Build a multimodal database to collect and integrate historical fault information of CNC machine tools. This information is stored in the database in the form of data, text, images, and videos, and categorized and annotated. Related literature on the fault information is collected and stored in the database. Based on a large language model, relevant fault query and maintenance suggestion functions are provided.

[0008] S2. Collecting and normalizing the working data of the CNC machine tool, wherein the working data includes vibration signals of the guide rail and the lead screw, vibration signals of the strain gauge, temperature field, and vibration signals of the spindle bearing and the rolling element bearing;

[0009] S3. Perform correlation analysis on the working data to generate feature vectors, wherein the feature vectors include feature vectors of the guide rail and the lead screw, feature vectors of the strain gauge, feature vectors of the temperature, and feature vectors of the spindle bearing and the rolling element bearing;

[0010] S4. Perform correlation analysis on the feature vectors to generate a fused feature vector and a fused feature matrix. Perform correlation analysis on the fused feature matrix to generate an equipment fault assessment index. Output the CNC machine tool status and maintenance recommendations based on the equipment fault assessment index.

[0011] Furthermore, historical fault information of CNC machine tools is collected, including guide rail and lead screw faults, strain gauge faults, servo motor faults, spindle bearing and rolling element bearing faults, and relevant data, text, pictures and videos of the corresponding faults are stored in a classified database.

[0012] Furthermore, relevant literature on fault information is collected and stored in the database. Based on the GPT and DeepSeek large language model, feature retrieval queries for corresponding faults are provided and integrated to provide maintenance suggestions for the corresponding faults.

[0013] Furthermore, the working data of the CNC machine tool is collected S times in one hour, with discrete collection and a sampling frequency of 1kHz, and the working data is normalized. The formula based on the normalization is:

[0014]

[0015] Where x(t) is the working data collected at sampling point t, min[x(t)] and max[x(t)] are the minimum and maximum values ​​of the corresponding historical working data when the CNC machine tool is working normally, and f(t) is the working data after normalization.

[0016] Guide rail and screw vibration signal v d (t) is the acceleration value at sampling point t, and the guide rail and screw vibration signal v d (t) is decomposed into 8 frequency bands, and the decomposed frequency bands and the guide rail and screw vibration signals v d (t) Perform correlation analysis to generate the energy E of the jth frequency band j , based on the formula:

[0017]

[0018] Among them, energy E j It is used to reflect the vibration intensity of the jth frequency band. The guide rail and screw vibration signal v is decomposed by wavelet packet. d (t) is decomposed into 3 layers, namely 2 3 frequency band, W j (n) is the coefficient of the jth frequency band;

[0019] Energy E j Perform correlation analysis to generate guide rail and lead screw feature vectors The formula is:

[0020] Furthermore, the strain gauge vibration signal is collected synchronously. The strain gauge vibration signal ε(t) is the acceleration value collected at the sampling point t. The strain gauge vibration signal ε(t) is subjected to correlation analysis to generate the signal mean μ, signal standard deviation σ, peak factor K, and skewness γ. The formula is as follows:

[0021]

[0022] Among them, the signal mean μ is used to reflect the average level of the strain gauge vibration signal ε(t), the signal standard deviation σ is used to reflect the fluctuation amplitude of the strain gauge vibration signal ε(t), the peak factor K is used to reflect the ratio of the maximum value of the strain gauge vibration signal ε(t) to the standard deviation, and is used to reflect the peak prominence of the strain gauge vibration signal, and the skewness γ is used to reflect the skewness of the distribution of the strain gauge vibration signal ε(t);

[0023] The strain gauge characteristic vector f is generated by integrating the signal mean μ, signal standard deviation σ, peak factor K, and skewness γ. ε , the formula is: f ε =[μ,σ,K,γ].

[0024] Furthermore, the motor temperature value is collected synchronously, and the maximum temperature value and the minimum temperature value are processed to generate the maximum temperature difference ΔT, and the collected motor temperature values ​​are averaged to generate the temperature average.

[0025] Comprehensive maximum temperature difference ΔT, temperature mean Generate temperature feature vector f T , based on the formula:

[0026] Furthermore, the vibration signals of the main shaft bearing and the rolling element bearing are collected synchronously, and the vibration signals v a (t) is the acceleration value collected at sampling point t, and the vibration signal v of the main shaft bearing and rolling element bearing is a (t) Calculate the Hilbert envelope e(t). The Hilbert envelope e(t) is generated by Hilbert transform according to the following formula:

[0027]

[0028] Among them, H[v a (t)] Vibration signal v of main shaft bearing and rolling element bearing a The Hilbert transform of (t) and the Hilbert envelope e(t) are used to reflect the instantaneous change of the vibration signals of the spindle bearing and the rolling element bearing;

[0029] Perform correlation analysis on the Hilbert envelope e(t) to generate the envelope mean μ e and kurtosis K e , kurtosis K e The formula is: Among them, σ e is the standard deviation of the Hilbert envelope e(t), max[e(t)] is the maximum value of the Hilbert envelope e(t), and the envelope mean μ e Used to reflect the signal mean of the Hilbert envelope e(t), kurtosis K e Used to reflect the prominence of the peak signal of the Hilbert envelope e(t);

[0030] Comprehensive envelope mean μ e and kurtosis K e , generate the spindle bearing and rolling element bearing feature vectors The formula is:

[0031] Furthermore, the correlation analysis of the feature vectors is performed to generate the fusion feature vector F s , based on the formula:

[0032]

[0033] in, ω ε 、ω T 、 They are the guide rail and screw fault weight factor, strain gauge fault weight factor, servo motor fault weight factor, spindle bearing and rolling element bearing fault weight factor, F s is the value of the fusion feature vector collected for the sth time, the subscript s is used to index the number of times, and generate the fusion feature matrix X, X = [F1, F2, F3, ..., F S ];

[0034] Perform correlation analysis on the fusion feature matrix X to generate the covariance matrix C based on the following formula: in, The mean vector of the fusion feature matrix X is selected, and the eigenvector u1 corresponding to the maximum eigenvalue is selected as the principal component direction. The corresponding eigenvector u1 and the fusion feature vector F are s Perform correlation analysis to generate equipment failure assessment index SPZ s , based on the formula:

[0035] SPZ s =F s u1

[0036] Equipment Failure Assessment Index SPZ s is a real number, which is used to reflect the maximum variance direction of the fusion feature vector obtained for the sth time, and is used to reflect the degree to which the equipment tends to be healthy. The maximum value of the equipment fault assessment index max(SPZ s ) is compared with the threshold θ, when max(SPZ s )≥θ, the equipment is in a healthy state and does not need to be repaired. s )<θ, the equipment is in an unhealthy state and requires manual inspection and maintenance.

[0037] The present invention provides a multimodal data fusion equipment evaluation and maintenance suggestion system for executing a multimodal data fusion equipment evaluation and maintenance suggestion method, comprising:

[0038] A multimodal database construction module is used to collect and integrate historical fault information of CNC machine tools, store and classify the fault information in the database in the form of data, text, pictures, and videos, collect relevant literature on the fault information and store it in the database, and provide relevant fault query and maintenance suggestion functions based on a large language model;

[0039] A data acquisition module is used to collect the working data of the CNC machine tool and normalize the working data. The working data includes the vibration signals of the guide rail and the lead screw, the vibration signals of the strain gauge, the temperature field, and the vibration signals of the spindle bearing and the rolling element bearing;

[0040] A data analysis module is used to perform correlation analysis on the working data and generate feature vectors, wherein the feature vectors include the guide rail and lead screw feature vectors, the strain gauge feature vectors, the temperature feature vectors, and the spindle bearing and rolling element bearing feature vectors;

[0041] The output module is used to perform correlation analysis on the feature vectors, generate fused feature vectors, and generate fused feature matrices, perform correlation analysis on the fused feature matrices, generate equipment fault assessment indexes, and output CNC machine tool status and maintenance recommendations based on the equipment fault assessment indexes.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] The present invention constructs a classification database containing multi-modal historical fault information, combines weight factors, and uses a large language model to realize fault feature retrieval and intelligent output of maintenance suggestions. It effectively integrates vibration signals and temperature fields, analyzes bearings, strain gauges, motors, guide rails and screws one by one, extracts feature vectors, and performs correlation analysis on the feature vectors. It integrates multi-source features to generate an equipment fault assessment index, and outputs the CNC machine tool status and maintenance suggestions based on the equipment fault assessment index. It improves the accuracy and sensitivity of fault diagnosis, can dynamically reflect the health status of the equipment, assists staff in taking maintenance measures in a timely manner, and significantly improves the safety of equipment operation and production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 Schematic diagram of the overall method flow of the present invention;

[0045] Figure 2 This is a module diagram of the overall system of the present invention. DETAILED DESCRIPTION

[0046] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to specific embodiments.

[0047] To further illustrate the technical means and effects employed by this application to achieve the intended invention objectives, the following detailed description of the specific implementation methods, structures, features, and effects of this application is provided in conjunction with the accompanying drawings and preferred embodiments. In the following description, different references to "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics of one or more embodiments may be combined in any suitable manner.

[0048] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0049] Example:

[0050] See also Figure 1 , the present invention provides a technical solution:

[0051] A multimodal data fusion equipment evaluation and maintenance recommendation method, comprising the following steps:

[0052] S1. Build a multimodal database to collect and integrate historical fault information of CNC machine tools. This information is stored in the database in the form of data, text, images, and videos, and categorized and annotated. Related literature on the fault information is collected and stored in the database. Based on a large language model, relevant fault query and maintenance suggestion functions are provided.

[0053] Collect historical fault information of CNC machine tools, including guide rail and lead screw faults, strain gauge faults, servo motor faults, and spindle bearing and rolling element bearing faults. Store relevant data, text, pictures, and videos of the corresponding faults in a classified database. CNC machine tool experts review and score the data to generate guide rail and lead screw fault weight factors, strain gauge fault weight factors, servo motor fault weight factors, and spindle bearing and rolling element bearing fault weight factors.

[0054] The database stores fault event tables, sensor data tables, text description tables, and image multimedia tables, and classifies them by fault event type. CNC machine tool experts score the severity and impact of the fault on a scale of 1-10. The larger the score, the more serious the fault. A comprehensive score is given to each type of screw fault, strain gauge fault, servo motor fault, spindle bearing, and rolling element bearing fault, and the values ​​are normalized. The fault weight factor is generated after weighted average, and the fault weight factor is scaled according to the proportion to meet the requirements. Experts refer to professionals who have profound professional knowledge, rich practical experience and the ability to solve complex problems in the field of CNC machine tools. They are able to comprehensively analyze data such as machine tool operating status, fault information, maintenance records, etc.

[0055] Relevant literature on fault information is collected and stored in a database. GPT and DeepSeek large language models are used to perform feature retrieval and provide integrated repair recommendations for the corresponding faults. Both GPT and DeepSeek are intelligent large language models that provide comprehensive feature analysis and repair recommendations based on relevant literature, making them convenient for staff to query.

[0056] S2. Collecting and normalizing the working data of the CNC machine tool, wherein the working data includes vibration signals of the guide rail and the lead screw, vibration signals of the strain gauge, temperature field, and vibration signals of the spindle bearing and the rolling element bearing;

[0057] S3. Perform correlation analysis on the working data to generate feature vectors, wherein the feature vectors include feature vectors of the guide rail and the lead screw, feature vectors of the strain gauge, feature vectors of the temperature, and feature vectors of the spindle bearing and the rolling element bearing;

[0058] The working data of the CNC machine tool is collected S times in one hour, with discrete collection and a sampling frequency of 1kHz. The working data is normalized. The formula for normalization is:

[0059]

[0060] Where x(t) is the working data collected at sampling point t, min[x(t)] and max[x(t)] are the minimum and maximum values ​​of the corresponding historical working data when the CNC machine tool is working normally, and f(t) is the working data after normalization.

[0061] Guide rail and screw vibration signal v d (t) is the acceleration value at sampling point t, and the guide rail and screw vibration signal v d (t) is decomposed into 8 frequency bands, and the decomposed frequency bands and the guide rail and screw vibration signals v d (t) Perform correlation analysis to generate the energy E of the jth frequency band j , based on the formula:

[0062]

[0063] Among them, energy E j It is used to reflect the vibration intensity of the jth frequency band. The guide rail and screw vibration signal v is decomposed by wavelet packet. d (t) is decomposed into 3 layers, namely 2 3 frequency band, W j (n) is the coefficient of the jth frequency band;

[0064] Energy E j Perform correlation analysis to generate guide rail and lead screw feature vectors The formula is:

[0065] Energy E j Reflects the dynamic state of the guide rail and screw. Abnormal energy concentration usually means mechanical looseness, unbalanced wear or assembly defects, and can indicate changes in the quality and rigidity of the mechanical guide rail movement.

[0066] The strain gauge vibration signal is collected synchronously. The strain gauge vibration signal ε(t) is the acceleration value collected at the sampling point t. The strain gauge vibration signal ε(t) is subjected to correlation analysis to generate the signal mean μ, signal standard deviation σ, peak factor K, and skewness γ. The formula is as follows:

[0067]

[0068] Among them, the signal mean μ is used to reflect the average level of the strain gauge vibration signal ε(t), the signal standard deviation σ is used to reflect the fluctuation amplitude of the strain gauge vibration signal ε(t), the peak factor K is used to reflect the ratio of the maximum value of the strain gauge vibration signal ε(t) to the standard deviation, and is used to reflect the peak prominence of the strain gauge vibration signal, and the skewness γ is used to reflect the skewness of the distribution of the strain gauge vibration signal ε(t);

[0069] The strain gauge characteristic vector f is generated by integrating the signal mean μ, signal standard deviation σ, peak factor K, and skewness γ. ε , the formula is: f ε =[μ,σ,K,γ].

[0070] The signal mean μ reflects the static trend, the signal standard deviation σ reflects the degree of structural vibration or micro-deformation, the peak factor K reflects the instantaneous extreme stress situation, and the skewness γ reflects the asymmetry of the signal distribution. The larger the value, the worse the working health of the strain gauge.

[0071] The motor temperature value is collected synchronously, and the maximum temperature value and the minimum temperature value are processed to generate the maximum temperature difference ΔT. The collected motor temperature value is averaged to generate the temperature average.

[0072] The maximum temperature difference ΔT reflects the unevenness of temperature distribution. The larger it is, the worse the working condition is.

[0073] Comprehensive maximum temperature difference ΔT, temperature mean Generate temperature feature vector f T , based on the formula:

[0074] Synchronously collect the vibration signals of the main shaft bearing and the rolling element bearing, the vibration signals of the main shaft bearing and the rolling element bearing v a (t) is the acceleration value collected at sampling point t, and the vibration signal v of the main shaft bearing and rolling element bearing is a (t) Calculate the Hilbert envelope e(t). The Hilbert envelope e(t) is generated by Hilbert transform according to the following formula:

[0075]

[0076] Among them, H[v a (t)] Vibration signal v of main shaft bearing and rolling element bearing a The Hilbert transform of (t) and the Hilbert envelope e(t) are used to reflect the instantaneous change of the vibration signals of the spindle bearing and the rolling element bearing;

[0077] Perform correlation analysis on the Hilbert envelope e(t) to generate the envelope mean μ e and kurtosis K e , kurtosis K e The formula is: Among them, σ e is the standard deviation of the Hilbert envelope e(t), max[e(t)] is the maximum value of the Hilbert envelope e(t), and the envelope mean μ eUsed to reflect the signal mean of the Hilbert envelope e(t), kurtosis K e Used to reflect the prominence of the peak signal of the Hilbert envelope e(t);

[0078] Comprehensive envelope mean μ e and kurtosis K e , generate the spindle bearing and rolling element bearing feature vectors The formula is: Characteristic vectors of spindle bearings and rolling element bearings Used to evaluate the health of bearings. The larger the value, the worse the health of the spindle bearings and rolling element bearings. It can help staff to promptly detect problems such as bearing rolling element defects, wear and insufficient lubrication.

[0079] S4. Perform correlation analysis on the feature vectors to generate a fused feature vector and a fused feature matrix. Perform correlation analysis on the fused feature matrix to generate an equipment fault assessment index. Output the CNC machine tool status and maintenance recommendations based on the equipment fault assessment index.

[0080] Furthermore, the correlation analysis of the feature vectors is performed to generate the fusion feature vector F s , based on the formula:

[0081]

[0082] in, ω ε 、ω T 、 They are the guide rail and screw fault weight factor, strain gauge fault weight factor, servo motor fault weight factor, spindle bearing and rolling element bearing fault weight factor, F s is the value of the fusion feature vector collected for the sth time, the subscript s is used to index the number of times, and generate the fusion feature matrix X, X = [F1, F2, F3, ..., F S The guide rail and screw eigenvectors, strain gauge eigenvectors, servo motor temperature eigenvectors, and spindle and rolling bearing vibration eigenvectors are each multiplied by their respective weighting factors and then accumulated. The weighting factors are designed to be non-negative, and all weights sum to 1. Therefore, the fused eigenvector is a weighted average of these different fault signatures. This process ensures that the fused eigenvector reflects the importance of each fault type while preventing a single fault signature from excessively influencing the overall assessment. This is logically sound and conforms to the principle of multi-source information fusion.

[0083] Perform correlation analysis on the fusion feature matrix X to generate the covariance matrix C based on the following formula: in, The eigenvector u1 corresponding to the maximum eigenvalue is selected as the principal component direction for the mean vector of the fused feature matrix X. The fused feature matrix is ​​composed of fused feature vectors from multiple acquisition moments, representing the dynamic state of the device over time. Correlation analysis is performed on this fused feature matrix, namely, calculating its covariance matrix to characterize the relationships between features and the overall variation of the data.

[0084] The covariance matrix is ​​calculated by averaging the matrix products of the fused feature data minus its mean. This process effectively removes mean bias and highlights the fluctuations in the data. Each element in the covariance matrix reflects the degree of linear correlation between two features. A positive value indicates that the two features increase or decrease simultaneously, i.e., positive correlation; a negative value indicates that the two features have opposite trends, i.e., negative correlation; and zero or near-zero indicates no significant linear relationship. Covariance analysis can identify the dominant direction of variation between features, providing a basis for subsequent principal component analysis.

[0085] For the corresponding eigenvector u1 and the fusion eigenvector F s Perform correlation analysis to generate equipment failure assessment index SPZ s , based on the formula:

[0086] SPZ s =F s u1

[0087] The eigenvector corresponding to the maximum eigenvalue of the covariance matrix is ​​selected as the principal component direction. This principal component is the direction with the greatest variance in the multidimensional feature space, meaning that projections along this direction maximize the discrete or changing characteristics of the data. The fused eigenvector is projected onto this principal component direction to obtain the device fault assessment index. The fault assessment index essentially reflects the value of the current device state in the direction of maximum variance; larger values ​​indicate a greater degree of deviation from the normal baseline. Since the principal component directions are defined in the space behind the data mean, the sign of the projected value indicates the direction of deviation of the device state from the mean. Larger positive values ​​correspond to a healthier or normal device state, while lower or negative values ​​correspond to an abnormal or faulty state. Therefore, the fault assessment index is positively correlated with the device health. In other words, a larger device fault assessment index indicates a healthier device, and vice versa.

[0088] Equipment Failure Assessment Index SPZ s is a real number, which is used to reflect the maximum variance direction of the fusion feature vector obtained for the sth time, and is used to reflect the degree to which the equipment tends to be healthy. The maximum value of the equipment fault assessment index max(SPZ s ) is compared with the threshold θ, which is obtained based on the statistical analysis of historical health status data. When max(SPZ s)≥θ, the equipment is in a healthy state and does not need to be repaired. s )<θ, the equipment is in an unhealthy state and requires manual inspection and maintenance.

[0089] Reference Figure 2 The present invention provides a multimodal data fusion equipment evaluation and maintenance suggestion system for executing a multimodal data fusion equipment evaluation and maintenance suggestion method, comprising:

[0090] A multimodal database construction module is used to collect and integrate historical fault information of CNC machine tools, store and classify the fault information in the database in the form of data, text, pictures, and videos, collect relevant literature on the fault information and store it in the database, and provide relevant fault query and maintenance suggestion functions based on a large language model;

[0091] A data acquisition module is used to collect the working data of the CNC machine tool and normalize the working data. The working data includes the vibration signals of the guide rail and the lead screw, the vibration signals of the strain gauge, the temperature field, and the vibration signals of the spindle bearing and the rolling element bearing;

[0092] A data analysis module is used to perform correlation analysis on the working data and generate feature vectors, wherein the feature vectors include the guide rail and lead screw feature vectors, the strain gauge feature vectors, the temperature feature vectors, and the spindle bearing and rolling element bearing feature vectors;

[0093] The output module is used to perform correlation analysis on the feature vectors, generate fused feature vectors, and generate fused feature matrices, perform correlation analysis on the fused feature matrices, generate equipment fault assessment indexes, and output CNC machine tool status and maintenance recommendations based on the equipment fault assessment indexes.

[0094] The above formulas are all dimensionless calculations. The formula is a formula that is closest to the actual situation obtained by software simulation of collected data. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0095] It should be noted that the order in which the embodiments of the present application are presented is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. Furthermore, the foregoing descriptions of specific embodiments of this specification are provided. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential sequence shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0096] The various embodiments in this application are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments.

[0097] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A multimodal data fusion equipment evaluation and maintenance recommendation method, characterized in that: The specific steps include: S1. Build a multimodal database to collect and integrate historical fault information of CNC machine tools. This information is stored in the database in the form of data, text, images, and videos, and categorized and annotated. Related literature on the fault information is collected and stored in the database. Based on a large language model, relevant fault query and maintenance suggestion functions are provided. S2. Collecting and normalizing the working data of the CNC machine tool, wherein the working data includes vibration signals of the guide rail and the lead screw, vibration signals of the strain gauge, temperature field, and vibration signals of the spindle bearing and the rolling element bearing; S3. Perform correlation analysis on the working data to generate feature vectors, wherein the feature vectors include feature vectors of the guide rail and the lead screw, feature vectors of the strain gauge, feature vectors of the temperature, and feature vectors of the spindle bearing and the rolling element bearing; S4. Perform correlation analysis on the feature vectors to generate a fused feature vector and a fused feature matrix. Perform correlation analysis on the fused feature matrix to generate an equipment fault assessment index. Output the CNC machine tool status and maintenance recommendations based on the equipment fault assessment index.

2. The multimodal data fusion equipment assessment and maintenance recommendation method according to claim 1, characterized in that: Collect historical fault information of CNC machine tools, including guide rail and lead screw faults, strain gauge faults, servo motor faults, spindle bearing and rolling element bearing faults, and store relevant data, text, pictures and videos of the corresponding faults in a classified database.

3. The multimodal data fusion equipment assessment and maintenance recommendation method according to claim 1, characterized in that: Collect relevant literature on fault information and store it in the database. Based on GPT and DeepSeek large language models, provide feature retrieval queries for corresponding faults and integrate and give maintenance suggestions for the corresponding faults.

4. The multimodal data fusion equipment assessment and maintenance recommendation method according to claim 1, characterized in that: The working data of the CNC machine tool is collected S times in one hour, with discrete collection and a sampling frequency of 1kHz. The working data is normalized. The formula for normalization is: Where x(t) is the working data collected at sampling point t, min[x(t)] and max[x(t)] are the minimum and maximum values ​​of the corresponding historical working data when the CNC machine tool is working normally, and f(t) is the working data after normalization. Guide rail and screw vibration signal v d (t) is the acceleration value at sampling point t, and the guide rail and screw vibration signal v d (t) is decomposed into 8 frequency bands, and the decomposed frequency bands and the guide rail and screw vibration signals v d (t) Perform correlation analysis to generate the energy E of the jth frequency band j , based on the formula: Among them, energy E j It is used to reflect the vibration intensity of the jth frequency band. The guide rail and screw vibration signal v is decomposed by wavelet packet. d (t) is decomposed into 3 layers, namely 2 3 frequency band, W j (n) is the coefficient of the jth frequency band; Energy E j Perform correlation analysis to generate guide rail and lead screw feature vectors The formula is:

5. The multimodal data fusion equipment evaluation and maintenance recommendation method according to claim 4, characterized in that: The strain gauge vibration signal is collected synchronously. The strain gauge vibration signal ε(t) is the acceleration value collected at the sampling point t. The strain gauge vibration signal ε(t) is subjected to correlation analysis to generate the signal mean μ, signal standard deviation σ, peak factor K, and skewness γ. The formula is as follows: Among them, the signal mean μ is used to reflect the average level of the strain gauge vibration signal ε(t), the signal standard deviation σ is used to reflect the fluctuation amplitude of the strain gauge vibration signal ε(t), the peak factor K is used to reflect the ratio of the maximum value of the strain gauge vibration signal ε(t) to the standard deviation, and is used to reflect the peak prominence of the strain gauge vibration signal, and the skewness γ is used to reflect the skewness of the distribution of the strain gauge vibration signal ε(t); The strain gauge characteristic vector f is generated by integrating the signal mean μ, signal standard deviation σ, peak factor K, and skewness γ. ε , the formula is: f ε =[μ,σ,K,γ].

6. The multimodal data fusion equipment assessment and maintenance recommendation method according to claim 5, characterized in that: The motor temperature value is collected synchronously, and the maximum temperature value and the minimum temperature value are processed to generate the maximum temperature difference ΔT. The collected motor temperature value is averaged to generate the temperature average. Comprehensive maximum temperature difference ΔT, temperature mean Generate temperature feature vector f T , based on the formula:

7. The multimodal data fusion equipment assessment and maintenance recommendation method according to claim 6, characterized in that: Synchronously collect the vibration signals of the main shaft bearing and the rolling element bearing, the vibration signals of the main shaft bearing and the rolling element bearing v a (t) is the acceleration value collected at sampling point t, and the vibration signal v of the main shaft bearing and rolling element bearing is a (t) Calculate the Hilbert envelope e(t). The Hilbert envelope e(t) is generated by Hilbert transform according to the following formula: Among them, H[v a (t)] Vibration signal v of main shaft bearing and rolling element bearing a The Hilbert transform of (t) and the Hilbert envelope e(t) are used to reflect the instantaneous change of the vibration signals of the spindle bearing and the rolling element bearing; Perform correlation analysis on the Hilbert envelope e(t) to generate the envelope mean μ e and kurtosis K e , kurtosis K e The formula is: Among them, σ e is the standard deviation of the Hilbert envelope e(t), max[e(t)] is the maximum value of the Hilbert envelope e(t), and the envelope mean μ e Used to reflect the signal mean of the Hilbert envelope e(t), kurtosis K e Used to reflect the prominence of the peak signal of the Hilbert envelope e(t); Comprehensive envelope mean μ e and kurtosis K e , generate the spindle bearing and rolling element bearing feature vectors The formula is:

8. The multimodal data fusion equipment evaluation and maintenance recommendation method according to claim 7, characterized in that: Perform correlation analysis on the feature vectors to generate the fusion feature vector F s , based on the formula: in, They are the guide rail and screw fault weight factor, strain gauge fault weight factor, servo motor fault weight factor, spindle bearing and rolling element bearing fault weight factor, F s is the value of the fusion feature vector collected for the sth time, the subscript s is used to index the number of times, and generate the fusion feature matrix X, X = [F1, F2, F3, ..., F S ]; Perform correlation analysis on the fusion feature matrix X to generate the covariance matrix C based on the following formula: in, The mean vector of the fusion feature matrix X is selected, and the eigenvector u1 corresponding to the maximum eigenvalue is selected as the principal component direction. The corresponding eigenvector u1 and the fusion feature vector F are s Perform correlation analysis to generate equipment failure assessment index SPZ s , based on the formula: License plate s =F s ·u1 Equipment Failure Assessment Index SPZ s is a real number, which is used to reflect the maximum variance direction of the fusion feature vector obtained for the sth time, and is used to reflect the degree to which the equipment tends to be healthy. The maximum value of the equipment fault assessment index max(SPZ s ) is compared with the threshold θ, when max(SPZ s )≥θ, the equipment is in a healthy state and does not need to be repaired. s )<θ, the equipment is in an unhealthy state and requires manual inspection and maintenance.

9. A multimodal data fusion equipment evaluation and maintenance suggestion system, used to execute the multimodal data fusion equipment evaluation and maintenance suggestion method according to claim 1, characterized in that: include: A multimodal database construction module is used to collect and integrate historical fault information of CNC machine tools, store and classify the fault information in the database in the form of data, text, pictures, and videos, collect relevant literature on the fault information and store it in the database, and provide relevant fault query and maintenance suggestion functions based on a large language model; A data acquisition module is used to collect the working data of the CNC machine tool and normalize the working data. The working data includes the vibration signals of the guide rail and the lead screw, the vibration signals of the strain gauge, the temperature field, and the vibration signals of the spindle bearing and the rolling element bearing; A data analysis module is used to perform correlation analysis on the working data and generate feature vectors, wherein the feature vectors include the guide rail and lead screw feature vectors, the strain gauge feature vectors, the temperature feature vectors, and the spindle bearing and rolling element bearing feature vectors; The output module is used to perform correlation analysis on the feature vectors, generate fused feature vectors, and generate fused feature matrices, perform correlation analysis on the fused feature matrices, generate equipment fault assessment indexes, and output CNC machine tool status and maintenance recommendations based on the equipment fault assessment indexes.

Images