Numerical control machine tool fault diagnosis and active early warning system

Abstract

The invention relates to the technical field of numerical control machine tool fault diagnosis, in particular to a numerical control machine tool fault diagnosis and active early warning system. The system comprises a data acquisition unit which is used for acquiring multi-dimensional original data generated in the operation process of the numerical control machine tool; the data processing and feature extraction unit performs feature extraction on the collected multi-dimensional original data based on an interactive feature construction algorithm; and the fault diagnosis unit constructs a classifier for distinguishing different fault types based on a random forest model, and realizes the judgment of the fault types through the random forest model based on the interactive feature vector # imgabs0 #. An interactive feature construction algorithm is adopted to perform interactive feature analysis on the preprocessed multi-dimensional original data, the problem of feature distortion caused by neglecting wear factors is avoided by introducing a real-time updated wear index, interference of external factors on feature analysis is eliminated by introducing an environment temperature correction item, and the accuracy of feature analysis is improved. And the actual reaction capability on the operation state of the equipment is improved.

Description

Technical Field

[0001] The invention relates to the technical field of numerical control machine tool fault diagnosis, and in particular to a numerical control machine tool fault diagnosis and active early warning system. Background Art

[0002] The CNC machine tool fault diagnosis and active warning system is an industrial equipment monitoring system based on multi-sensor data acquisition, feature analysis and intelligent algorithms. It is used to monitor the operating status of CNC machine tools in real time, diagnose potential faults, and provide active warnings before faults occur. It collects and analyzes multi-dimensional signals and combines intelligent algorithms to achieve real-time monitoring of equipment operating status, accurate identification of fault modes and active warning functions.

[0003] In CNC machine tools that operate at high loads for a long time, mechanical wear will cause performance changes. However, in the existing CNC machine fault diagnosis systems, the performance changes caused by mechanical wear during long-term operation of the equipment are often ignored, resulting in distortion of the fault diagnosis model features and reducing the accuracy of fault detection. Therefore, a CNC machine fault diagnosis and active early warning system is designed. Summary of the invention

[0004] The purpose of the present invention is to provide a CNC machine tool fault diagnosis and active warning system to solve the problem raised in the above background technology that performance changes caused by mechanical wear during long-term operation lead to distortion of fault diagnosis model features and reduce the accuracy of fault detection.

[0005] To achieve the above object, the present invention aims to provide a fault diagnosis and active early warning system for a numerically controlled machine tool, comprising a data acquisition unit, wherein the data acquisition unit is used to collect multi-dimensional raw data generated during the operation of the numerically controlled machine tool;

[0006] Among them, the multi-dimensional raw data includes vibration signals, temperature signals, current signals, cutting force signals and historical fault data;

[0007] The data processing and feature extraction unit extracts features from the collected multi-dimensional raw data based on the interactive feature construction algorithm to obtain a preliminary interactive feature vector. , and based on the multi-input multi-output model, the interactive feature vector is constructed and optimized ;

[0008] A fault diagnosis unit is constructed based on a random forest model to distinguish different fault types. The fault type can be judged through the random forest model;

[0009] An active warning unit sends an active warning signal based on the result determined by the fault diagnosis unit.

[0010] As a further improvement of the technical solution, the data acquisition unit includes a vibration monitoring module, a temperature monitoring module, a current monitoring module and a cutting force monitoring module;

[0011] Among them, the vibration monitoring module is used to monitor vibration signals and capture abnormal conditions of mechanical parts;

[0012] The temperature monitoring module is used to monitor the temperature signal;

[0013] The current monitoring module is used to monitor the current signal, load fluctuation and abnormal current;

[0014] The cutting force monitoring module is used to monitor the cutting force signal and the cutting force changes.

[0015] As a further improvement of the technical solution, the data processing and feature extraction unit includes a data processing module and a feature extraction module;

[0016] Among them, the data processing module is used to pre-process the collected multi-dimensional raw data;

[0017] The feature extraction module performs interactive feature analysis on the preprocessed multidimensional raw data based on the interactive feature construction algorithm, constructs a preliminary interactive feature vector, and finally outputs the optimized interactive feature vector through a multi-input multi-output model.

[0018] As a further improvement of the technical solution, the specific steps involved in the interactive feature construction algorithm for interactive feature analysis of multi-dimensional raw data are:

[0019] S1.1. Calculate the intensity ratio of vibration signal to cutting force signal:

[0020] ;

[0021] in, ; ;

[0022] In the formula, It represents the ratio between the vibration signal energy and the cutting force signal energy, and is used to reflect the energy coupling characteristics of the two signals; Indicates the energy of vibration signal, which is used to reflect the vibration intensity when the machine tool is running; Indicates the energy of the cutting force signal, which is used to reflect the magnitude of the cutting force during machine tool processing; The vibration signal Sampling point values; Indicates the total number of sampling points of the vibration signal; Represents an index variable; The cutting force signal Sampling point values; Represents the total number of sampling points of the cutting force signal;

[0023] Considering that the wear of machine tools will increase the vibration of mechanical parts, the vibration signal intensity ratio after wear correction is constructed. :

[0024] ;

[0025] In the formula, is the adjustment coefficient, which indicates the influence of wear on the vibration signal; It represents the wear index, reflecting the aging and performance degradation trend of mechanical parts over time;

[0026] S1.2. Current increase is usually accompanied by temperature rise. Calculate the slope characteristics of the temperature change rate and the current mean:

[0027] ;

[0028] in, ; ;

[0029] In the formula, It represents the ratio of temperature change rate to current mean value; It represents the rate of temperature change, which indicates the rate of temperature change over time; Represents the mean value of the current signal; Represents the total number of sampling points of the current signal; The current signal Sampling point values; Indicates the temperature value at the end of sampling; Indicates the temperature value at the beginning of sampling; Indicates the sampling time interval;

[0030] Considering the influence of ambient temperature on the temperature change rate, construct the temperature change rate feature considering the ambient temperature change :

[0031] ;

[0032] In the formula, Indicates the degree of influence of environmental conditions on the rate of temperature change; Indicates the ambient temperature;

[0033] S1.3. Calculate the weighted combination features of multi-dimensional raw data:

[0034] ;

[0035] In the formula, represents the weighted combination feature; represents the mean value of the temperature signal; The weight representing the energy of the vibration signal; The weight representing the mean value of the temperature signal; The weight representing the mean value of the current signal; Indicates the peak value of the cutting force signal; represents the weight of the peak value of the cutting force signal;

[0036] S1.4. Define the preliminary interaction feature vector :

[0037] ;

[0038] in, ;

[0039] In the formula, It represents the power spectrum density of the vibration signal, reflecting the energy distribution of the signal in the frequency domain; It represents the variance of the current signal and the fluctuation amplitude of the signal.

[0040] As a further improvement of the technical solution, the multi-input multi-output model performs the preliminary interaction feature vector The specific steps involved in the optimization are:

[0041] S2.1. Time series data and the preliminary interaction feature vector As input signal of multi-input multi-output model;

[0042] The input time series data Specifically:

[0043] ;

[0044] In the formula, Indicates vibration signal; Indicates temperature signal; Indicates current signal; Indicates the cutting force signal; Represents the input time series data;

[0045] Then the time series of vibration, temperature, current and cutting force signals are:

[0046] ;

[0047] In the formula, Indicates input signal For different signal sources (vibration, temperature, current and cutting force), the time series of distinguish; Indicates input signal exist The value at each sampling moment is used to describe the temporal variation of the signal; Indicates the sampling time point;

[0048] S2.2. For each time series signal Use one-dimensional convolution to extract features:

[0049] ;

[0050] In the formula, represents the convolution kernel; Represents the bias used in the convolution operation; Represents the convolution operation; Represents convolutional features;

[0051] At the same time, the preliminary feature vector Mapped to high-dimensional space through a fully connected layer:

[0052] ;

[0053] In the formula, represents the weight of the fully connected layer; Represents the bias vector of the fully connected layer; Represents the initial interaction feature vector from the fully connected layer The high-dimensional feature vector generated after mapping to the high-dimensional space; represents the activation function;

[0054] S2.3. Convolutional features and interaction features To perform the fusion:

[0055] ;

[0056] In the formula, Represents feature splicing; Represents the fused feature vector, which is used for the subsequent generation of optimized feature vectors; Indicates vibration signal The feature vector extracted by one-dimensional convolution; Indicates temperature signal The feature vector extracted by one-dimensional convolution; Indicates current signal The feature vector extracted by one-dimensional convolution; Indicates cutting force signal The feature vector extracted by one-dimensional convolution; Represents the initial interaction feature vector The interactive feature vector generated after being mapped to a high-dimensional space through a fully connected layer;

[0057] S2.4. Fusion features Output the optimized feature vector through the fully connected layer :

[0058] ;

[0059] in, Represents the fully connected layer operation. The fully connected layer operation is specifically ;in, Represents the optimization of the output fully connected layer weights;

[0060] Optimized interaction feature vector ;

[0061] In the formula, Indicates An optimized feature.

[0062] As a further improvement of the technical solution, the specific steps of constructing the random forest model classifier are:

[0063] Collect historical data through the data acquisition unit, including samples of normal operation and various fault conditions;

[0064] The optimized interaction feature vector and the corresponding fault label As the original training set ;

[0065] During training, each decision tree is trained by bootstrapping from the original training set. Generate data subsets by random sampling , the original training set Include Sample ,in, Indicates The feature vector of the samples, Indicates the corresponding label;

[0066] In the process of training a decision tree, each time a node is split, a random Extract a feature subset from , and in this feature subset Select the best split point in

[0067] Each decision tree Based on different feature subsets and a subset of the data Conduct training;

[0068] Repeat the above process to generate decision trees to form a random forest model ;

[0069] Building a classifier using the random forest algorithm :

[0070] ;

[0071] In the formula, Represents a given optimized feature vector Finally, the classification results output by the random forest model represent the fault type of the machine tool; represents the number of decision trees; Indicates The prediction results of a decision tree; Represents the majority operation.

[0072] As a further improvement of the technical solution, the specific steps of the random forest model for fault type judgment are:

[0073] The optimized interactive feature vector generated in real time by the running CNC machine tool As input features;

[0074] Each decision tree Independent prediction of failure categories :

[0075] ;

[0076] In the formula, Represents all possible fault categories; Indicates the total number of fault categories;

[0077] The prediction results of multiple decision trees are combined into the final category through a voting mechanism;

[0078] ;

[0079] In the formula, Represents the final prediction result of the random forest model; Represents the number of decision trees in the random forest; It means taking the majority of all decision tree prediction results, that is, the category with the most votes;

[0080] Compute the probability that an input feature belongs to each fault class:

[0081] ;

[0082] In the formula, Represents the feature vector Belongs to the fault category probability; represents the indicator function, Indicates when The prediction results of the decision tree Equal to category The value is 1 when , otherwise 0; Indicates that in all decision trees, the fault category The predicted frequency, thus obtaining the probability of the category; Represents an index variable;

[0083] Based on the voting results and probability output, determine the fault type:

[0084]

[0085] In the formula, represents the final predicted fault category; Indicates the fault type probability.

[0086] As a further improvement of the technical solution, the active warning unit includes a warning module and a fault output module; the warning module is based on the final predicted fault category and active warning rules to issue active warning signals;

[0087] Among them, the active warning rules are as follows:

[0088] Based on fault category The degree of harm is divided into different warning levels, including ordinary, serious and urgent;

[0089] For each fault category Preset the corresponding threshold, if the probability of the fault type If the set threshold is exceeded, the warning module triggers an alert;

[0090] The fault output module generates a fault report by outputting the warning signal, warning level and fault category information that triggers the warning module and stores it in the database.

[0091] Compared with the prior art, the present invention has the following beneficial effects:

[0092] 1. In this CNC machine tool fault diagnosis and active warning system, an interactive feature construction algorithm is used to perform interactive feature analysis on the preprocessed multi-dimensional raw data to construct a preliminary interactive feature vector; wherein, by introducing a real-time updated wear index, the model can dynamically respond to changes in equipment status and avoid feature distortion problems caused by ignoring wear factors. At the same time, the introduction of an ambient temperature correction term helps to eliminate the interference of external factors on feature analysis and improve the actual reflection ability of the equipment operating status.

[0093] 2. In this CNC machine tool fault diagnosis and active warning system, a multi-input multi-output model is used to process time series data (such as vibration, temperature, current, cutting force, etc.) from different sensors, and the local features of each signal are extracted through one-dimensional convolution, so as to more comprehensively capture the operating status of the system; by fusing convolution features and interaction features, the multi-input multi-output model can capture the complex interaction effects between different features, such as wear-vibration coupling features and environment-temperature coupling features, which helps to more accurately identify potential fault modes. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] Figure 1 It is a flowchart of the overall process of the present invention.

[0095] The meaning of each number in the figure is:

[0096] 1. Data acquisition unit; 2. Data processing and feature extraction unit; 21. Data processing module; 22. Feature extraction module; 3. Fault diagnosis unit; 4. Active warning unit; 41. Warning module; 42. Fault output module. DETAILED DESCRIPTION

[0097] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0098] Example

[0099] See also Figure 1 As shown, a numerical control machine tool fault diagnosis and active early warning system is provided, including a data acquisition unit 1, the data acquisition unit 1 is used to collect multi-dimensional raw data generated during the operation of the numerical control machine tool;

[0100] Among them, the multi-dimensional raw data includes vibration signals, temperature signals, current signals, cutting force signals and historical fault data;

[0101] In this embodiment, the data acquisition unit 1 includes a vibration monitoring module, a temperature monitoring module, a current monitoring module and a cutting force monitoring module;

[0102] Among them, the vibration monitoring module is installed on the main shaft and bearings to monitor the vibration signal and capture the abnormal status of the mechanical parts;

[0103] The temperature monitoring module is installed on the spindle, bearings and motor to monitor the temperature signal;

[0104] The current monitoring module is installed on the motor and the drive to monitor the current signal, load fluctuation and abnormal current;

[0105] The cutting force monitoring module is installed on the cutting tool to monitor the cutting force signal and the cutting force change.

[0106] The vibration monitoring module is preferably a vibration sensor: at the spindle and bearing, it is used to monitor vibration signals and capture abnormal conditions of mechanical parts;

[0107] The temperature monitoring module is preferably a temperature sensor: at the spindle, bearing, and motor, used to monitor temperature signals and detect overheating problems;

[0108] The current monitoring module is preferably a current sensor: motor and driver, used to monitor current signals, load fluctuations and abnormal current;

[0109] The cutting force monitoring module is preferably a force sensor: a cutting tool, which is used to monitor the cutting force signal and monitor the cutting force changes.

[0110] This embodiment also includes a data processing and feature extraction unit 2, which extracts features from the collected multi-dimensional raw data based on an interactive feature construction algorithm to obtain a preliminary interactive feature vector , and based on the multi-input multi-output model, the interactive feature vector is constructed and optimized ;

[0111] Specifically, the data processing and feature extraction unit 2 includes a data processing module 21 and a feature extraction module 22;

[0112] The data processing module 21 is used to pre-process the collected multi-dimensional raw data;

[0113] The vibration, temperature, current and cutting force signals are collected in the form of time series, and the original multi-dimensional time series signal set is:

[0114] ;

[0115] In the formula, Indicates A set of data points of a signal; represents the collected original multi-dimensional time series signal set;

[0116] Use a low-pass filter to remove high-frequency noise:

[0117] ;

[0118] In the formula, Represents the signal obtained after low-pass filtering; Represents the original time series signal; Represents the impulse response function of the filter, which is used to define the characteristics of the low-pass filter; Indicates the time offset;

[0119] And perform signal normalization to eliminate the influence of different dimensions on feature extraction:

[0120] ;

[0121] In the formula, Represents the normalized signal data point, which is used to eliminate the influence between different dimensions; Indicates the value of the current signal data point; represents the signal mean; represents standard deviation;

[0122] The feature extraction module 22 performs interactive feature analysis on the preprocessed multi-dimensional raw data based on an interactive feature construction algorithm, constructs a preliminary interactive feature vector, and finally outputs an optimized interactive feature vector through a multi-input multi-output model.

[0123] In this embodiment, the specific steps involved in the interactive feature construction algorithm for performing interactive feature analysis on multi-dimensional raw data are:

[0124] S1.1. The vibration intensity increases with the cutting force. The intensity ratio of the vibration signal to the cutting force signal is calculated:

[0125] ;

[0126] in, ; ;

[0127] In the formula, It represents the ratio between the vibration signal energy and the cutting force signal energy, and is used to reflect the energy coupling characteristics of the two signals; Indicates the energy of vibration signal, which is used to reflect the vibration intensity when the machine tool is running; Indicates the energy of the cutting force signal, which is used to reflect the magnitude of the cutting force during machine tool processing; The vibration signal Sampling point values; Indicates the total number of sampling points of the vibration signal; Represents an index variable; The cutting force signal Sampling point values; Represents the total number of sampling points of the cutting force signal;

[0128] Considering that the wear of machine tools will increase the vibration of mechanical parts, the vibration signal intensity ratio after wear correction is constructed. :

[0129] ;

[0130] In the formula, is the adjustment coefficient, which indicates the influence of wear on the vibration signal; It represents the wear index, reflecting the aging and performance degradation trend of mechanical parts over time;

[0131] S1.2. Current increase is usually accompanied by temperature rise. Calculate the slope characteristics of the temperature change rate and the current mean:

[0132] ;

[0133] in, ; ;

[0134] In the formula, It represents the ratio of temperature change rate to current mean value; It represents the rate of temperature change, which indicates the rate of temperature change over time; Represents the mean value of the current signal; Represents the total number of sampling points of the current signal; The current signal Sampling point values; Indicates the temperature value at the end of sampling; Indicates the temperature value at the beginning of sampling; Indicates the sampling time interval;

[0135] Considering the influence of ambient temperature on the temperature change rate, construct the temperature change rate feature considering the ambient temperature change :

[0136] ;

[0137] In the formula, Indicates the degree of influence of environmental conditions on the rate of temperature change; Indicates the ambient temperature;

[0138] S1.3. Calculate the weighted combination features of multi-dimensional raw data:

[0139] ;

[0140] In the formula, represents the weighted combination feature; represents the mean value of the temperature signal; The weight representing the energy of the vibration signal; The weight representing the mean value of the temperature signal; The weight representing the mean value of the current signal; Indicates the peak value of the cutting force signal; represents the weight of the peak value of the cutting force signal;

[0141] Weighted combination features can highlight the correlation between various signals and avoid the one-sidedness that may be caused by isolating and analyzing a single signal. Multidimensional data contains a large number of features, and directly processing these features may increase the computational complexity. Weighted combination features merge multiple features into one indicator, which plays a role in dimensionality reduction.

[0142] S1.4. Define the preliminary interaction feature vector :

[0143] ;

[0144] in, ;

[0145] In the formula, It represents the power spectrum density of the vibration signal, reflecting the energy distribution of the signal in the frequency domain; Represents the variance of the current signal, indicating the fluctuation amplitude of the signal;

[0146] Preliminary interaction feature vector Processing and fusing information from multi-dimensional raw data (such as vibration, temperature, current and cutting force) and extracting the interactive characteristics between these signals (such as energy coupling, signal correlation, dynamic change trend, etc.) can help reveal the potential physical correlation between multi-dimensional signals, such as the energy ratio of vibration to cutting force, the relationship between temperature change and current, etc., reflecting the inherent laws of the equipment's operating status; helping to reveal the potential physical correlation between multi-dimensional signals, such as the energy ratio of vibration to cutting force, the relationship between temperature change and current, etc., reflecting the inherent laws of the equipment's operating status; at the same time, the original signal usually contains a large number of sampling points, with a high dimension and more redundant information. By defining the preliminary interactive feature vector, the feature dimension can be effectively reduced while retaining important information, thereby reducing data complexity;

[0147] Specifically, the multi-input multi-output model performs the initial interaction feature vector The specific steps involved in the optimization are:

[0148] S2.1. Time series data and the preliminary interaction feature vector As input signal of multi-input multi-output model;

[0149] The input time series data Specifically:

[0150] ;

[0151] In the formula, Indicates vibration signal; Indicates temperature signal; Indicates current signal; Indicates the cutting force signal; Represents the input time series data;

[0152] Then the time series of vibration, temperature, current and cutting force signals are:

[0153] ;

[0154] In the formula, Indicates input signal For different signal sources (vibration, temperature, current and cutting force), the time series of distinguish; Indicates input signal exist The value at each sampling moment is used to describe the temporal variation of the signal; Indicates the sampling time point;

[0155] In the machine tool fault diagnosis and active early warning system, there are the following common signal sources:

[0156] Vibration signal Used to reflect the vibration of key machine tool components (such as spindles, bearings, etc.) during operation, and monitor abnormal vibration, mechanical resonance or imbalance problems of bearings or spindles;

[0157] Vibration signal The time series is:

[0158] ;

[0159] In the formula, Indicates that the vibration signal is The amplitude of the sampling points;

[0160] Temperature signal Used to reflect the temperature changes of key machine tool components (such as spindle motors, bearings or cutting tools) and monitor whether motors, tools, bearings and other components are overheated;

[0161] Temperature signal The time series is:

[0162] ;

[0163] In the formula, Indicates that the temperature signal is The amplitude of the sampling points;

[0164] Current signal Reflects the current fluctuation of the drive motor or control circuit, and is used to monitor the status of the electrical system (load fluctuation of the drive motor, motor overload or control circuit abnormality);

[0165] Current signal The time series is:

[0166] ;

[0167] In the formula, Indicates that the current signal is The amplitude of the sampling points;

[0168] Cutting force signal It is used to reflect the cutting force on the tool during the machining process, monitor the force of tool wear or breakage during the machining process, and the changes in the hardness of the machining material or the clamping force;

[0169] Cutting force signal The time series is:

[0170]

[0171] In the formula, Indicates the cutting force signal The amplitude of the sampling points;

[0172] S2.2. For each time series signal Use one-dimensional convolution to extract features:

[0173] ;

[0174] In the formula, represents the convolution kernel; Represents the bias used in the convolution operation; Represents the convolution operation; Represents convolutional features;

[0175] At the same time, the preliminary feature vector Mapped to high-dimensional space through a fully connected layer:

[0176] ;

[0177] In the formula, represents the weight of the fully connected layer; Represents the bias vector of the fully connected layer; Represents the initial interaction feature vector from the fully connected layer The high-dimensional feature vector generated after mapping to the high-dimensional space; represents the activation function;

[0178] S2.3. Convolutional features and interaction features To perform the fusion:

[0179] ;

[0180] In the formula, Represents feature splicing; Represents the fused feature vector, which is used for the subsequent generation of optimized feature vectors; Indicates vibration signal The feature vector extracted by one-dimensional convolution; Indicates temperature signal The feature vector extracted by one-dimensional convolution; Indicates current signal The feature vector extracted by one-dimensional convolution; Indicates cutting force signal The feature vector extracted by one-dimensional convolution; Represents the initial interaction feature vector The interactive feature vector generated after being mapped to a high-dimensional space through a fully connected layer;

[0181] S2.4. Fusion features Output the optimized feature vector through the fully connected layer :

[0182] ;

[0183] in, Represents the fully connected layer operation. The fully connected layer operation is specifically ;in, Represents the optimization of the output fully connected layer weights;

[0184] Optimized interaction feature vector ;

[0185] In the formula, Indicates An optimized feature.

[0186] Specifically, the training process of the multi-input multi-output model is:

[0187] Time series data and the initial feature vector As model data input;

[0188] The output of the multi-input multi-output model is the optimized feature vector ;

[0189] When fault classification is performed through a multi-input multi-output model, the loss function is:

[0190] ;

[0191] When fault prediction is performed using a multi-input multi-output model, the loss function is:

[0192] ;

[0193] In the formula, represents the total number of samples; Indicates the number of categories; and All represent index variables; Representation sample The one-hot encoded value of the actual category label; Representation sample Category The predicted probability of Represents samples in continuous value prediction tasks The true value of Represents the model prediction output value in the continuous value prediction task; Represents the classification task loss function; Represents the regression task loss function;

[0194] Optimizing weights through back-propagation algorithm , , ; Specifically, by calculating the loss function relative to each weight ( , , ), the back propagation algorithm updates these weights to reduce the loss, which can make the model more accurate in each link of feature extraction, interactive feature learning and final output prediction, thereby improving the performance of the model and completing the fault classification and prediction tasks.

[0195] The multi-input multi-output model can simultaneously process time series data from different sensors (such as vibration, temperature, current, cutting force, etc.), and extract the local features of each signal through one-dimensional convolution, so as to more comprehensively capture the operating status of the system; by fusing convolution features and interaction features, the multi-input multi-output model can capture the complex interaction effects between different features, such as wear-vibration coupling features and environment-temperature coupling features, which helps to more accurately identify potential failure modes; by introducing real-time updated wear index and environmental conditions, the model can dynamically respond to changes in equipment status, adjust the warning threshold in time, and reduce false alarm and missed alarm rates.

[0196] This embodiment also includes a fault diagnosis unit 3, which constructs a classifier for distinguishing different fault types based on a random forest model and an interactive feature vector. The fault type can be judged through the random forest model;

[0197] Among them, the specific steps of constructing the random forest model classifier are:

[0198] Collect historical data through the data acquisition unit, including samples of normal operation and various fault conditions;

[0199] The optimized interaction feature vector and the corresponding fault label As the original training set ;

[0200] During training, each decision tree is trained by bootstrapping from the original training set. Generate data subsets by random sampling , the original training set Include Sample ,in, Indicates The feature vector of the samples, Indicates the corresponding label;

[0201] In the process of training a decision tree, each time a node is split, a random Extract a feature subset from , and in this feature subset Select the best split point in

[0202] Each decision tree Based on different feature subsets and a subset of the data Conduct training;

[0203] Repeat the above process to generate decision trees to form a random forest model ;

[0204] Building a classifier using the random forest algorithm :

[0205] ;

[0206] In the formula, Represents a given optimized feature vector Finally, the classification results output by the random forest model represent the fault type of the machine tool; represents the number of decision trees; Indicates The prediction results of a decision tree; Represents the majority operation, that is, all decision trees The voting results of the predicted results.

[0207] Furthermore, the specific steps of the random forest model for fault type judgment are:

[0208] The optimized interactive feature vector generated in real time by the running CNC machine tool As input features;

[0209] Each decision tree Independent prediction of failure categories :

[0210] ;

[0211] In the formula, Represents all possible fault categories; Indicates the total number of fault categories;

[0212] The prediction results of multiple decision trees are combined into the final category through a voting mechanism;

[0213] ;

[0214] In the formula, Indicates the final prediction result of the random forest model, which is the fault category obtained by voting on the prediction results of multiple decision trees. The voting mechanism determines the final fault category by voting. The system determines the category with the most votes as that category. Represents the number of decision trees in the random forest; It means taking the majority of all decision tree prediction results, that is, the category with the most votes;

[0215] Compute the probability that an input feature belongs to each fault class:

[0216] ;

[0217] In the formula, Represents the feature vector Belongs to the fault category probability; represents the indicator function, Indicates when The prediction results of the decision tree Equal to category The value is 1 when , otherwise 0; Indicates that in all decision trees, the fault category The predicted frequency, thus obtaining the probability of the category; Represents an index variable;

[0218] Based on the voting results and probability output, determine the fault type:

[0219]

[0220] In the formula, represents the final predicted fault category; Indicates the fault type probability.

[0221] Specifically, identify all possible fault types and assign a unique label or number to each fault type; for each input feature vector , the random forest model gives the probability of belonging to each fault type , if the probability of the fault type If the set threshold is exceeded, an alarm for this fault type is triggered;

[0222] If the probability of two or more categories exceeds their respective thresholds, all eligible fault types are recorded. , and trigger alarms for these fault types simultaneously;

[0223] Each decision tree For a given feature vector Each decision tree will independently predict a fault category, and the prediction results of all decision trees will be counted as the final prediction result. .

[0224] In this embodiment, an active warning unit 4 is also included, and the active warning unit 4 sends an active warning signal based on the result determined by the fault diagnosis unit 3;

[0225] The active warning unit 4 includes a warning module 41 and a fault output module 42; the warning module 41 is based on the final predicted fault category. and active warning rules to issue active warning signals;

[0226] Among them, the specific active warning rules are:

[0227] Based on fault category The degree of harm is divided into different warning levels, including ordinary, serious and urgent;

[0228] For each fault category Preset the corresponding threshold, if the probability of the fault type If the set threshold is exceeded, the warning module 41 triggers a warning;

[0229] The fault output module 42 generates a fault report by outputting the warning signal, warning level and fault category information that triggers the warning of the warning module 41 and stores it in the database.

[0230] The fault report shall include at least:

[0231] Fault type: such as bearing failure, motor overheating, current fluctuation, etc.;

[0232] Affected components: such as spindle, motor, transmission system, etc.

[0233] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and the above embodiments and descriptions are only preferred examples of the present invention, and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. A fault diagnosis and active early warning system for CNC machine tools, characterized in that: include A data acquisition unit (1), the data acquisition unit (1) being used to acquire multi-dimensional raw data generated during the operation of a numerically controlled machine tool; Among them, the multi-dimensional raw data includes vibration signals, temperature signals, current signals, cutting force signals and historical fault data; A data processing and feature extraction unit (2), wherein the data processing and feature extraction unit (2) extracts features from the collected multi-dimensional raw data based on an interactive feature construction algorithm to obtain a preliminary interactive feature vector , and based on the multi-input multi-output model, the interactive feature vector is constructed and optimized ; A fault diagnosis unit (3) is constructed based on a random forest model to construct a classifier for distinguishing different fault types, based on an interactive feature vector The fault type can be judged through the random forest model; An active warning unit (4), wherein the active warning unit (4) sends an active warning signal based on a result determined by the fault diagnosis unit (3).

2. The CNC machine tool fault diagnosis and active early warning system according to claim 1, characterized in that: The data acquisition unit (1) comprises a vibration monitoring module, a temperature monitoring module, a current monitoring module and a cutting force monitoring module; Among them, the vibration monitoring module is used to monitor vibration signals and capture abnormal conditions of mechanical parts; The temperature monitoring module is used to monitor the temperature signal; The current monitoring module is used to monitor the current signal, load fluctuation and abnormal current; The cutting force monitoring module is used to monitor the cutting force signal and the cutting force changes.

3. The CNC machine tool fault diagnosis and active early warning system according to claim 2 is characterized in that: The data processing and feature extraction unit (2) comprises a data processing module (21) and a feature extraction module (22); Wherein, the data processing module (21) is used to pre-process the collected multi-dimensional raw data; The feature extraction module (22) performs interactive feature analysis on the pre-processed multi-dimensional raw data based on an interactive feature construction algorithm, constructs a preliminary interactive feature vector, and finally outputs an optimized interactive feature vector through a multi-input multi-output model.

4. The CNC machine tool fault diagnosis and active early warning system according to claim 3 is characterized in that: The specific steps involved in the interactive feature construction algorithm for interactive feature analysis of multi-dimensional raw data are: S1.

1. Calculate the intensity ratio of vibration signal to cutting force signal: ； In the formula, It represents the ratio between the vibration signal energy and the cutting force signal energy; Represents the energy of vibration signal; Indicates the energy of the cutting force signal; Considering that the wear of machine tools will increase the vibration of mechanical parts, the vibration signal intensity ratio after wear correction is constructed. : ； In the formula, is the adjustment coefficient, which indicates the influence of wear on the vibration signal; Indicates the wear index; S1.

2. Calculate the slope characteristics of the temperature change rate and the current mean: ； In the formula, It represents the ratio of temperature change rate to current mean value; Indicates the rate of temperature change; Represents the mean value of the current signal; Considering the influence of ambient temperature on the temperature change rate, construct the temperature change rate feature considering the ambient temperature change : ； In the formula, Indicates the degree of influence of environmental conditions on the rate of temperature change; Indicates the ambient temperature; S1.

3. Calculate the weighted combination features of multi-dimensional raw data: ； In the formula, represents the weighted combination feature; represents the mean value of the temperature signal; The weight representing the energy of the vibration signal; The weight representing the mean value of the temperature signal; The weight representing the mean value of the current signal; Indicates the peak value of the cutting force signal; represents the weight of the peak value of the cutting force signal; S1.

4. Define the preliminary interaction feature vector : ； In the formula, Represents the power spectral density of the vibration signal; Represents the variance of the current signal.

5. The CNC machine tool fault diagnosis and active early warning system according to claim 4 is characterized in that: The multi-input multi-output model is used to calculate the initial interaction feature vector The specific steps involved in the optimization are: S2.

1. Time series data and the preliminary interaction feature vector As input signal of multi-input multi-output model; The input time series data Specifically: ； In the formula, Indicates vibration signal; Indicates temperature signal; Indicates current signal; Indicates the cutting force signal; Represents the input time series data; Then the time series of vibration, temperature, current and cutting force signals are: ； In the formula, Indicates input signal time series; Indicates input signal exist The value at each sampling moment; Indicates the sampling time point; S2.

2. For each time series signal Use one-dimensional convolution to extract features: ； In the formula, represents the convolution kernel; Represents the bias used in the convolution operation; Represents the convolution operation; Represents convolutional features; At the same time, the preliminary feature vector Mapped to high-dimensional space through a fully connected layer: ； In the formula, represents the weight of the fully connected layer; Represents the bias vector of the fully connected layer; Represents the initial interaction feature vector from the fully connected layer The high-dimensional feature vector generated after mapping to the high-dimensional space; represents the activation function; S2.

3. Convolutional features and interaction features To perform the fusion: ； In the formula, Represents feature splicing; Represents the fused feature vector, which is used for the subsequent generation of optimized feature vectors; Indicates vibration signal The feature vector extracted by one-dimensional convolution; Indicates temperature signal The feature vector extracted by one-dimensional convolution; Indicates current signal The feature vector extracted by one-dimensional convolution; Indicates cutting force signal The feature vector extracted by one-dimensional convolution; Represents the initial interaction feature vector The interactive feature vector generated after being mapped to a high-dimensional space through a fully connected layer; S2.

4. Fusion features Output the optimized feature vector through the fully connected layer : ； in, Represents the fully connected layer operation; Optimized interaction feature vector ; In the formula, Indicates An optimized feature.

6. The CNC machine tool fault diagnosis and active early warning system according to claim 5, characterized in that: The specific steps of constructing the random forest model classifier are: Collect historical data through the data acquisition unit, including samples of normal operation and various fault conditions; The optimized interaction feature vector and the corresponding fault label As the original training set ; During training, each decision tree is trained by bootstrapping from the original training set. Random sampling to generate data subsets , the original training set Include Sample ,in, Indicates The feature vector of the samples, Indicates the corresponding label; In the process of training a decision tree, each time a node is split, a random Extract a feature subset from , and in this feature subset Select the best split point in Each decision tree Based on different feature subsets and a subset of the data Conduct training; Repeat the above process to generate decision trees to form a random forest model ; Building a classifier using the random forest algorithm : ； In the formula, Represents a given optimized feature vector Finally, the classification results output by the random forest model; represents the number of decision trees; Indicates The prediction results of a decision tree; Represents the majority operation.

7. The CNC machine tool fault diagnosis and active early warning system according to claim 6 is characterized in that: The specific steps of the random forest model for fault type judgment are: The optimized interactive feature vector generated in real time by the running CNC machine tool As input features; Each decision tree Independent prediction of failure categories : ； In the formula, Represents all possible fault categories; Indicates the total number of fault categories; The prediction results of multiple decision trees are combined into the final category through a voting mechanism; ； In the formula, Represents the final prediction result of the random forest model; Represents the number of decision trees in the random forest; Indicates taking the mode of all decision tree prediction results; Compute the probability that an input feature belongs to each fault class: ； In the formula, Represents the feature vector Belongs to the fault category The probability of represents the indicator function; Indicates that in all decision trees, the fault category The frequency of prediction; Represents an index variable; Based on the voting results and probability output, determine the fault type: In the formula, represents the final predicted fault category; Indicates the fault type probability.

8. The CNC machine tool fault diagnosis and active early warning system according to claim 7, characterized in that: The active warning unit (4) comprises a warning module (41) and a fault output module (42); the warning module (41) outputs a fault type based on the final prediction. and active warning rules to issue active warning signals; Among them, the active warning rules are as follows: Based on fault category The degree of hazard is divided into different warning levels; For each fault category Preset the corresponding threshold, if the probability of the fault type If the set threshold is exceeded, the warning module (41) triggers a warning; The fault output module (42) outputs the warning signal, warning level and fault category information that triggers the warning of the warning module (41) to generate a fault report and stores it in a database.

Images