CN115716217A - Spindle runout abnormality detection method, spindle runout abnormality detection device, and storage medium - Google Patents
Spindle runout abnormality detection method, spindle runout abnormality detection device, and storage medium Download PDFInfo
- Publication number
- CN115716217A CN115716217A CN202211364856.8A CN202211364856A CN115716217A CN 115716217 A CN115716217 A CN 115716217A CN 202211364856 A CN202211364856 A CN 202211364856A CN 115716217 A CN115716217 A CN 115716217A
- Authority
- CN
- China
- Prior art keywords
- training sample
- deflection
- main shaft
- vibration data
- spindle
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The embodiment of the application provides a method and a device for detecting abnormal deflection of a main shaft and a storage medium, and relates to the technical field of computer digital control; acquiring a vibration data sample set of spindles of a plurality of CNC machine tables in a preset time period of multiple autorotation and the deflection level of each spindle; acquiring vibration data sample sets of main shafts of different types of CNC machine tables in a multiple-autorotation preset time period and the deflection radius of each main shaft; extracting a time domain characteristic set in a vibration data sample set to form a training sample set, wherein the training sample set comprises a plurality of training sample units, and each training sample unit comprises a time domain characteristic corresponding to vibration data and a corresponding deflection radius; performing model training on the training sample set by using a gradient lifting tree algorithm to obtain a prediction classification model; collecting real-time vibration data of the main shaft in real time; and loading the time domain characteristics corresponding to the real-time vibration data to a prediction classification model so as to predict the deflection level of the main shaft.
Description
[ technical field ] A method for producing a semiconductor device
The embodiment of the application relates to the technical field of computer digital control, in particular to a method and a device for detecting spindle deflection abnormity and a storage medium.
[ background of the invention ]
The traditional method for detecting the abnormality of the computer digital control equipment comprises the following steps: after the defective products quantity of CNC board production obviously increased, stopped the board operation, dismantled the cutter, stick and amesdial are examined in the installation, and the stick is examined in manual rotation, and the amesdial detected data is looked over to the human eye, expends time, and after the detected value exceeded the scope, the clearance main shaft and examined the sediment bits between the stick, then detect repeatedly, and measuring result all has the error at every turn, and is up to confirming the true scope that exceeds of detected value, lets the engineer maintain.
The traditional spindle deflection detection is only carried out when a large number of defective products are produced by a machine table, so that the yield of the products is reduced, extra cost is increased, and the reject ratio of the products is improved because the equipment is operated for a period of time under a bad condition.
[ summary of the invention ]
The embodiment of the application provides a method and a device for detecting spindle deflection abnormity and a storage medium, which are used for automatically detecting a spindle deflection angle according to CNC spindle vibration data acquired in real time, finding out the spindle deflection abnormity in advance, ensuring that equipment runs under a normal condition and reducing the labor maintenance cost.
In a first aspect, an embodiment of the present application provides a method for detecting spindle yaw abnormality, which is used for diagnosing and warning a spindle of a CNC machine. Collecting vibration data sample sets of a plurality of main shafts of CNC machine tables of different types in a preset time period of multiple autorotation and the deflection radius of each main shaft;
extracting a time domain feature set in the vibration data sample set to form a training sample set, wherein the training sample set comprises a plurality of training sample units, and the training sample units comprise time domain features corresponding to the vibration data and corresponding deflection radiuses;
performing model training on the training sample set by using a gradient lifting tree algorithm to obtain a prediction classification model, wherein the model training comprises the following steps: inputting the training sample unit into a model determined based on a loss function and the gradient lifting tree algorithm, performing iteration of a preset turn to update model parameters, and determining the prediction classification model according to the model parameters; collecting real-time vibration data of the main shaft in real time, and extracting corresponding time domain characteristics; loading the time domain characteristics corresponding to the real-time vibration data to the prediction classification model to obtain the current deflection radius of the main shaft, so as to predict the deflection level of the main shaft
The spindle deflection is detected by predicting the healthy state classification of the spindle deflection of the CNC machine. Collecting CNC machine main shaft vibration data with different health levels and deflection levels of all main shafts to serve as a training sample set, and performing model training on the training sample set by using a gradient lifting tree algorithm to obtain a prediction classification model; in the model application stage, acquiring vibration data of the CNC spindle in real time and loading the vibration data to a prediction classification model, periodically predicting the deflection level of the CNC spindle by the prediction classification model according to the real-time vibration data of the CNC spindle, and detecting whether the spindle deflection is abnormal or not according to the change condition of the deflection level; the main shaft deflection is timely and accurately found to be abnormal, the equipment is ensured to run under the normal condition, and the manpower maintenance cost is reduced.
In one possible implementation manner, the step of inputting the training sample unit into a model determined based on a loss function and the gradient lifting tree algorithm, and performing a preset iteration to update model parameters includes: constructing a decision tree for the time domain features based on the gradient lifting tree algorithm; forming a learner f based on the loss function through the iteration of t-1 t-1 (x) Wherein T is an iteration round number, T is more than or equal to 2 and less than or equal to T, T is a preset round number, and x is a feature element in the time domain feature in the training sample unit; after the t-th iteration, when the preset condition is met and the decision tree is established, the learner f t-1 (x) To form a learner f (x) and to update the model parameters.
In one possible implementation manner, after the t-th iteration, when a preset condition is met and the decision tree is completely built, the learner f performs the above-mentioned operation t-1 (x) The step of updating on the basis of (a) to form the learner f (x) includes: loss function pair learning device f based on current training sample unit to current t round number t-1 (x) Solving a first derivative and a second derivative, substituting the first derivative and the second derivative into a loss function to obtain a value optVs of the current loss function (t is based on the attempt of splitting a decision tree again by the current node, wherein the decision tree comprises a plurality of leaf nodes, the default current score is 0, the current score is compared with a value corresponding to the current loss function, the leaf nodes with larger values further split sub-trees, when the larger value is 0, the decision tree is completely established, and the learner f is t-1 (x) Is updated to form the learner f (x). In one possible implementation manner, the step of splitting the decision tree further includes: randomly selecting s features from the time domain features in the current training sample unitAnd the elements are arranged according to the sequence from small to large and are placed into the left sub-tree and the right sub-tree of the decision tree to split the decision tree.
In one possible implementation manner, the step of performing a preset iteration to update the model parameters further includes: optimizing the preset turns using a stochastic search algorithm to optimize the model parameters.
In one possible implementation, the time-domain feature includes at least one of a peak factor, a pulse factor, a margin factor, a waveform factor, a kurtosis factor, and a skewness factor of the vibration data.
In one possible implementation manner, the step of determining the prediction classification model according to the model parameter further includes: performing K-fold cross validation on the training sample set, dividing the training sample set into K parts, selecting K-1 parts for training each time, and taking the rest parts as a test set to obtain the prediction classification model; and selecting K-1 parts again to be used as a training set, and using the rest parts as a testing set to optimize the prediction classification model.
In one possible implementation manner, the method further includes: and when the deflection level of the current time based on the predicted deflection radius is lower than the deflection level of the main shaft before the current time, judging that the main shaft deflection is abnormal, and giving an early warning instruction to the CNC machine. In a second aspect, an embodiment of the present application provides a device for detecting a spindle runout abnormality, which is used for diagnosing and warning a spindle of a CNC machine, and the device includes:
the sensor is arranged on the main shaft and used for collecting a vibration data sample set of the main shaft in a preset time period of multiple autorotation;
a memory for storing a plurality of program modules;
a processor coupled to the sensor and reading a memory;
the processor is configured to load the plurality of program modules and execute the method for detecting spindle yaw anomaly provided by the method of the first aspect.
In a third aspect, an embodiment of the present application provides a computer-readable storage medium storing computer instructions, which cause the computer to execute the method provided in the first aspect.
It should be understood that the second to third aspects of the embodiment of the present application are consistent with the technical solution of the first aspect of the embodiment of the present application, and beneficial effects achieved by the aspects and the corresponding possible implementation are similar, and are not described again.
[ description of the drawings ]
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present specification, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without creative efforts.
Fig. 1 is a schematic diagram illustrating an abnormal spindle runout state of a CNC machine according to an embodiment of the present application;
FIG. 2 is a flowchart illustrating steps of a method for detecting spindle yaw anomaly according to an embodiment of the present disclosure;
FIG. 3 is a flowchart illustrating another method for detecting spindle yaw anomalies according to an embodiment of the present disclosure;
FIG. 4 is a diagram illustrating a comparison of decision trees according to the importance of feature elements in the present embodiment;
fig. 5 is a schematic structural diagram of a spindle runout abnormality detection apparatus according to an embodiment of the present application.
[ detailed description ] embodiments
For better understanding of the technical solutions in the present specification, the following detailed description of the embodiments of the present application is provided with reference to the accompanying drawings.
It should be understood that the described embodiments are only a few embodiments of the present specification, and not all embodiments. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments in the present specification without any creative effort belong to the protection scope of the present specification.
The terminology used in the embodiments of the present application is for the purpose of describing particular embodiments only and is not intended to be limiting of the specification. As used in the examples of this application and the appended claims, the singular forms "a", "an", and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.
Fig. 1 is a schematic view illustrating an abnormal state of spindle runout of a CNC machine provided in this embodiment of the application, and in a normal case, an axis of a spindle ZH is located in a vertical direction. As the machining time increases, the main shaft ZH deviates from the vertical direction by a certain angle
The main axis ZH ' is shown in the figure, and the main axis ZH ', the deflected main axis ZH ' and the angle
Within the determined triangle, the angle
Corresponding to the runout radius R. The health state of this application to the beat radius R of main shaft is carried out the grade classification to the main shaft, and when judging the healthy grade of board and descend, then give the suggestion of reporting to the police to make things convenient for follow-up overhaul of the equipments.
The method for detecting the spindle deflection abnormality in the embodiment of the application can be applied to electronic equipment or servers, such as computer clusters, terminals, cloud computers and the like, wherein the electronic equipment or servers are connected or in communication connection with a CNC machine.
Fig. 2 is a flowchart of steps of a method for detecting spindle yaw abnormality according to an embodiment of the present application, and as shown in fig. 2, the steps include:
step S21: the method comprises the steps of collecting vibration data sample sets of spindles of a plurality of different types of CNC machine tables in a preset time period of multiple autorotation and the deflection radius of each spindle.
In the embodiment, the types of the CNC machine are divided into three types, the corresponding main shaft deflection levels are respectively A type, B type or NG type, wherein the radius range of the main shaft deflection corresponding to the A type is 0mm to 0.02mm; the deflection radius range of the corresponding main shaft of the B class is 0.02-0.03 mm; the corresponding main shaft deflection radius of the NG class is more than 0.03mm. The method comprises the steps of collecting vibration data generated within a preset time period of main shaft autorotation of three different CNC machine tables, for example, within 15 seconds, simultaneously recording the deflection radius of a main shaft corresponding to different vibration data, and marking the level of main shaft deflection corresponding to the deflection radius of the main shaft corresponding to the collected vibration data.
In this embodiment, the vibration data is measured by an acceleration sensor provided on the outer surface or inside of the spindle. It is understood that the values measured by the sensor for 15 seconds are a series of vibration data, each of which includes values of vibration data in three directions. Therefore, the acquired data of the spindle of the CNC machine table of different types in 15 seconds is a plurality of groups of vibration data and forms a vibration data sample set. In this embodiment, each of the machines acquires 30 sets of data to form a vibration data sample set.
In other embodiments, the CNC machine can be further classified into more or different categories according to the main shaft deflection radius and the deflection precision. And the preset time period can also be 5-30 seconds and the like.
Step S22: extracting a time domain feature set in the vibration data sample set to form a training sample set, wherein the training sample set comprises a plurality of training sample units, and each training sample unit comprises a group of time domain features corresponding to the vibration data and a corresponding beat radius level.
In one example of the present application, the time-domain characteristic includes at least one of a peak factor, a pulse factor, a margin factor, a form factor, a kurtosis factor, and a skewness factor of the vibration data. Extracting the time domain feature set in the vibration data sample set may be a process of performing feature engineering operations to convert raw data into features that better express the nature of the problem, such that applying these features to a prediction model may improve the model prediction accuracy for invisible data. In the feature engineering, the following parameters of the following time domain features can be calculated for the vibration data: standard deviation, root mean square, skewness, etc.
Exemplarily, a 15-second vibration acceleration of the autorotation of a class A machine spindle, a 15-second vibration acceleration of the autorotation of a class B machine spindle, a 15-second vibration acceleration of the autorotation of a class C machine spindle are collected, time domain characteristics of the 15-second vibration acceleration of the autorotation of the class A machine spindle, the 15-second vibration acceleration of the autorotation of the class A machine spindle and the 15-second vibration acceleration of the autorotation of the class A machine spindle are respectively extracted, and the time domain characteristics and the class A deflection radius of the class A machine spindle are correspondingly stored to obtain a training sample unit 1; correspondingly storing the time domain characteristics of the main shaft of the B-class machine table and the B-class deflection radius to obtain a training sample unit 2; and correspondingly storing the time domain characteristics of the main shaft of the C-type machine and the C-type deflection radius to obtain a training sample unit 3.
Step S23: and carrying out model training on the training sample set by using a gradient lifting tree algorithm to obtain a prediction classification model.
The model training comprises the following steps: inputting the training sample unit into a model determined based on a loss function and the gradient lifting tree algorithm, performing iteration of a preset turn to update model parameters, and determining the prediction classification model according to the model parameters.
The embodiment of the application also provides that: inputting the training sample unit into a model determined based on a loss function and the gradient lifting tree algorithm, and performing a preset iteration to update model parameters, wherein the step comprises the following steps:
step S23-1: constructing a decision tree for the time domain features based on the gradient lifting tree algorithm;
exemplarily, assuming that the time domain features include a time domain feature a, a time domain feature B, and a time domain feature C, constructing a first decision tree based on the time domain feature a, constructing a second decision tree based on the time domain feature B, constructing a third decision tree based on the time domain feature C, and combining the first decision tree, the second decision tree, and the third decision tree to obtain an aggregate model, where the aggregate model summarizes prediction results of multiple decision trees.
Step S23-2: forming a learner f based on the loss function through the iteration of t-1 t-1 (x) Wherein T is an iteration round number, T is more than or equal to 2 and less than or equal to T, T is a preset round number, and x is a feature element in the time domain feature in the training sample unit;
the feature cells represent different parameters in the time domain features, and include time domain feature standard deviation, time domain feature root mean square, time domain feature skewness and the like obtained by calculating vibration data.
In different iteration rounds, different decision trees are split, and the predicted value of the total k decision trees of the decision tree model to the training sample set = the predicted value of the previous k-1 prediction trees + the predicted value of the kth tree.
Illustratively, the t-2 th iteration brings the root mean square of the temporal feature B into the learner f t-2 (x) Predicting the deflection radius of the main shaft as B according to the root mean square of the time domain feature B; in the iteration of the t-1, the root mean square of the time domain feature A is taken into a learner f t-1 (x) And predicting the deflection radius of the main shaft as a according to the root mean square of the time domain feature B, wherein the prediction result of the t-1 iteration is a + B.
Step S23-3: after the t-th iteration, when a preset condition is met and the decision tree is built, the learner f t-1 (x) To form a learner f (x) and to update the model parameters.
After the t-th iteration, when the preset condition is met and the decision tree is established, the learner f t-1 (x) The step of updating on the basis to form the learner f (x) includes:
loss function pair learning device f based on current training sample unit to current t round number t-1 (x) Solving a first derivative and a second derivative, and substituting the first derivative and the second derivative into a loss function to obtain the value of the current loss function;
attempting to split the decision tree again based on the current node, wherein the decision tree comprises a plurality of leaf nodes, the default current score is 0, the current score is compared with a value corresponding to the current loss function, the leaf nodes with larger score are further split into subtrees, when the larger value is 0, the decision tree is completely built, and the learner f is provided with a function for splitting the decision tree again t-1 (x) Is updated to form the learner f (x).
Constructing a model to be trained comprising a learning device, inputting a training sample set, the maximum falling times, a loss function and a regularization coefficient into the model to be trained, carrying out multiple iterations on the learning device by the model to be trained according to a gradient lifting tree algorithm, updating the loss function in the multiple iteration process, calculating a first derivative and a second derivative of the sample based on the learning device in the current iteration loss function, and calculating the first derivative and the second derivative of the sample; and according to the loss function and the first derivative and the second derivative of the sample, performing split tree judgment on the node of the current iteration, trying to update a larger score, and completing gradient promotion of the learner by correspondingly dividing the characteristic and splitting the sub-tree to execute parameter updating of the model based on the larger score.
In this embodiment, the step of performing model training on the training sample set by using the gradient lifting tree algorithm further includes: performing K-fold Cross Validation (also called cycle Validation) on the training sample set, dividing the training sample set into K parts, selecting K-1 parts for performing the training set each time, and taking the rest parts as a test set to obtain optimized model parameters; and (4) selecting K-1 parts again to be used as a training set, using the rest parts as a testing set, and circulating the steps so as to optimize the model parameters and further obtain an excellent prediction classification model. The K-fold cross validation can prevent the problem of overfitting of model parameters. In the training stage, model training is carried out on the collected training sample set by using a gradient lifting tree algorithm to obtain a prediction classification model. And in the application stage, the deviation angle of the main shaft of the CNC machine table during work is monitored in real time by using the prediction classification model.
Step S24: and acquiring real-time vibration data of the main shaft in real time, and extracting corresponding time domain characteristics.
For example, the machine to be diagnosed may collect 15 seconds of spindle rotation vibration data in real time, load the collected 15 seconds of vibration data operation parameters into a feature engineering for preprocessing, and extract corresponding time domain features, so as to ensure consistency of the real-time data and the training sample data set.
Step S25: and loading the time domain characteristics corresponding to the real-time vibration data to the prediction classification model to obtain the level of the deflection radius of the current main shaft so as to determine the deflection level of the main shaft.
The method for detecting the spindle deflection abnormity provided by the embodiment of the application collects the vibration data of spindles with different deflection levels, takes the time domain characteristics corresponding to a group of vibration data and the corresponding deflection levels as training sample units, and a training sample set is formed by a plurality of training sample units; and performing model training by adopting a gradient lifting tree algorithm according to the training sample set to obtain a prediction classification model for finishing classification of main shaft deflection according to vibration data, acquiring real-time vibration data in the operation process of the CNC machine equipment, loading the real-time vibration data into the prediction classification model to obtain the deflection radius of the main shaft of the CNC machine equipment in operation so as to determine the deflection level of the main shaft, wherein the process does not need manual participation, can predict whether the deflection of the main shaft is abnormal in advance, and reduces the damage caused by operation of the equipment under adverse conditions.
One embodiment of the application provides that the prediction classification model obtains a current predicted deflection level of the spindle, and can also compare the current predicted deflection level with a historical predicted deflection level, judge whether the level of the spindle deflection radius predicted at the current time is lower than the deflection radius level of the spindle before the current time, and when the level of the predicted deflection radius at the current time is lower than the deflection radius level of the spindle before the current time, judge that the spindle deflection is abnormal, and give an early warning instruction to the CNC machine; can carry out timely early warning to bad operation CNC equipment, send the early warning when the main shaft is healthy to slide and maintain in advance, reduce the defective products and damage the equipment performance condition and take place the probability.
Another embodiment of the present application provides that, in the process of training the preset classification model, the preset round may be optimized using a random search algorithm to optimize the preset classification model parameters. For data with a large enough sample size in the embodiment, the random search algorithm can also find a globally optimal preset round value at a high probability through random sampling, and the efficiency of parameter optimization can be improved by adopting the algorithm.
Another embodiment of the present application provides a specific implementation method for constructing a model including a learner, inputting the training sample unit to a loss function for performing the preset iteration to update the prediction classification model parameters, including the steps of:
step S231: and constructing a decision tree for the time domain features based on the gradient lifting tree algorithm.
The gradient lifting tree algorithm (LightGBM) is a tree model that combines many weak classifiers together to form a strong classifier. Each weak classifier is a tree structure, and the leaf nodes of each tree structure represent a score, not the class to which the sample belongs. The gradient lifting tree algorithm realizes the construction of a tree through continuous splitting characteristics. Since the prediction of the spindle runout essentially classifies the runout radius, a loss function of classification is employed when using the gradient lifting tree algorithm.
S232: forming a learner f through the iteration of t-1 times t-1 (x) And T is an iteration round number, T is more than or equal to 2 and less than or equal to T, and T is a preset round number and is one feature element in the time domain features in the training sample unit.
Learning device f t-1 (x) The first t-2 model predictions plus the t-1 tree predictions.
Step S233, after the t-th iteration, based on the loss function pair learner f of the current training sample unit to the current t rounds t-1 (x) And solving the first derivative and the second derivative, and substituting the first derivative and the second derivative into the loss function to obtain the value of the current loss function.
And S234, trying to split the decision tree again based on the current node, defaulting the current score to be 0, comparing the current score with a value corresponding to the current loss function, further splitting sub-trees by the leaf nodes with larger score, and finishing the establishment of the decision tree when the larger value is 0.
Step S235 in the learner f t-1 (x) To form a learner f (x) and to update the model parameters. When the model parameters are determined, the function of the prediction classification model is determined. The main shaft deflection type can be predicted according to the function of the prediction classification model.
The method comprises the steps of carrying out iteration minimization loss of a loss function for t times by a learner, calculating a first derivative and a second derivative of a sample based on the first derivative and the second derivative of the learner in each iteration calculation, arranging the samples from small to large according to characteristics, calculating the sum of first derivatives of a left sub-tree and a right sub-tree after each sample is placed in the left sub-tree in sequence according to the sequence of size arrangement, calculating the sum of second derivatives of the left sub-tree and the right sub-tree, updating scores according to the sum of the first derivatives of the left sub-tree and the right sub-tree and the sum of the second derivatives of the left sub-tree and the right sub-tree, dividing the characteristic parameters and the characteristic values of the split sub-trees according to the models, and determining whether to execute next split decision tree operation or not according to the scores.
In each iteration, a first derivative and a second derivative of a sample are brought into a learner which is output in the last iteration to obtain a target loss function in the current iteration, loss values in the calculation of the target loss function in the current iteration are marked according to the predicted values and the corresponding deflection radius levels of the vibration data in the training sample unit, namely the learner which is output in the current iteration updates scores, and model parameters are adjusted according to the scores.
In an example of performing the present invention to obtain a current loss function, the loss function is a function that maps an event (an element in a sample space) to a real number expressing the opportunity cost associated with its event, and is used to measure the degree of disparity between the predicted value f (x) and the true value Y of the model, thereby visually representing the association of some "cost" with the event. One goal of the optimization problem is to minimize the loss function. And the objective function is a loss function that is minimized under constrained conditions.
The objective function of the prediction classification model in this embodiment includes two parts, one part is a loss function l, which represents the difference between the predicted value and the true value, Ω is the regular term of the model,
the objective function is:
wherein the content of the first and second substances,
Simultaneously ordering:
wherein, g i Is the first derivative of the loss function, h i Is the second derivative of the loss function;
in the penalty terms, gamma and lambda represent regular term coefficients, namely penalty coefficients, and the gamma and lambda control the complexity of the tree structure of the decision tree; t represents the number of leaf nodes for a given tree, w represents the square of the output score on each leaf node; and calculating to obtain:
and then ordering:
wherein G is j As weighted first derivative of the loss function (i.e. asSum of first derivative), G j Is a weighted second derivative of the loss function (i.e., the sum of the second derivatives);
bringing in
And (3) pushing out:
g is to be j 、H j And substituting an equation to obtain a minimized loss function obj corresponding to the final prediction classification model:
the embodiment of the application is right in the process that the training sample set utilizes a gradient lifting tree algorithm to carry out model training to obtain a prediction classification model, the node with the largest splitting profit is selected from all current leaf nodes to be split, the splitting is carried out in a recursive manner, the overfitting is easy, the maximum depth needs to be limited because the node is easy to get into a higher depth, and the overfitting is avoided.
To further illustrate the method of fig. 3 by building a decision tree and iteratively updating model parameters, an embodiment of the flow of the method is further described below. In the above method
The input of the predictive classification model is the training sample unit I = { (x) 1 ，y 1 )，(x 2 ，y 2 )，......(x m ，y m ) The maximum iteration time T, a loss function L and a regularization coefficient; wherein x m Is a certain feature element in the time domain features of the vibration data in a group of collected training sample units, m is the format of the feature element, y m Is the corresponding yaw radius; the regularization coefficients (penalty coefficients) are gamma and lambda, and the output is a strong learner f (x);
t =1,2,3.
K11: calculating the i-th sample at the current round loss function l based on f t-1 (x i ) First derivative g ti Second derivative h ti And weighted first derivative G of all samples t And weighted second derivative H t 。
K12: based on the current node to try to split the decision tree, a default score =0,G is the sum of first derivatives of nodes needing to be classified currently, H is the sum of second derivatives of the nodes needing to be classified currently, K feature elements are randomly taken from time domain features of a training sample unit I, and the feature elements are arranged from small to large according to a sequence number K =1,2. Since the function of the predictive classification model is a minimized loss function, let
G L ＝0，H L ＝0；
Sequentially taking out the ith sample, and sequentially calculating the sum of the first derivative and the first derivative of the left subtree and the sum of the first derivative and the first derivative of the right subtree after the current sample is put into the left subtree:
G L ＝G L +g ti ，G R ＝G-G L ；
H L ＝H L +h ti ，H R ＝H-H L ；
mixing the above G L 、H L Substituting the above derived predictive classification model function obj, trying to update the maximum score:
k13: and splitting the subtrees based on the corresponding division features and the feature values of the maximum score.
K14: if the maximum score is 0, then the current decision tree is builtImmediately after completion, calculating w of all leaf regions tj Obtaining a strong learner f obtained by the current iteration weak learner t (x) As a weak learner for the next iteration; if the maximum score is not 0, proceed to K12 to attempt to split the decision tree.
It should be noted that, when splitting a decision tree, there is a predictor, i.e. a learner f (x), for each decision tree; and so on until all learners (decision trees) are done. Finally, the values in each decision tree are added to obtain the final prediction result, i.e. the strong learner f t (x) .1. The Learning device f t (x) Any of the previous learners is referred to as a weak learner.
Wherein the strong learning device
，
After the tree is split in this embodiment, the formed decision tree is sequentially split according to the importance degree of the feature elements in the time domain feature.
Fig. 4 shows a comparison chart of the decision tree of the present embodiment according to the importance degree of the feature element. Wherein the abscissa represents the importance degree of the feature elements to the prediction result, and the ordinate represents each feature element in the time domain feature. For example, mean _ vz represents the Mean of the principal axis z-direction runout at a certain time, and MaiChong _ vy represents the impulse factor of the runout in the y-direction.
The embodiment of the application also provides that after t iterations are performed on a plurality of training sample units, K-fold cross validation is performed, the sample data is randomly divided into K parts, K-1 parts are randomly selected as a training set each time, the rest part is used as a test set, and after the round is completed, K-1 parts are selected again to iterate the training data, so that the model accuracy is improved.
The method for detecting the abnormal deflection of the main shaft comprises the following steps:
k21: acquiring vibration data of a large number of equipment autorotations, namely acquiring vibration acceleration data of a machine table main shaft for 15 seconds of autorotation for machine table equipment of different deflection radius grades based on sensors arranged on standard machine tables of different deflection radius grades; the yaw radius classes include: (A: 0.02mm/B: 0.02-0.03 mm/NG:0.03 mm-).
K22: feature engineering-transforming data into corresponding time domain features.
K23: distinguishing a cross validation data set, carrying out K-fold cross validation on data, randomly dividing sample data into K parts, randomly selecting K-1 parts as a training set each time, using the rest part as a test set, and reselecting K-1 parts to iterate training data after the round is finished so as to improve the model accuracy.
K24: lightGBM builds a training model-model training is performed using the LightGBM machine learning algorithm.
K25: adjusting parameters using a random parameter optimization algorithm-searching for optimal LightGBM model parameters using a random search algorithm.
K26: optimal prediction model-the optimal LightGBM prediction model is built using the optimal model parameters.
K27: and (3) collecting vibration data of the diagnostic equipment in real time, namely collecting 15-second main shaft rotation vibration data of the machine to be diagnosed in real time.
K28: and (3) real-time data preprocessing, namely loading the acquired 15-second vibration main shaft operation parameters into a characteristic engineering preprocessing model so as to ensure the consistency of a training data set and the real-time data.
K29: judging the level of the main shaft deflection radius of the equipment by the prediction classification model, loading the preprocessed data characteristics into the prediction classification model, and judging the A/B/NG equipment by the prediction classification model.
K30: and detecting whether the main shaft deflection radius grade change-real-time diagnosis model result slides from the A-type equipment to the B/NG-type equipment.
K31: and sending an early warning signal, and sending the early warning signal when detecting that the deviation of the main shaft of the CNC machine equipment is predicted to fall from the A class to the B/NG class, and dispatching equipment maintenance personnel to carry out on-site abnormal troubleshooting.
Fig. 5 is a schematic structural diagram of a device for detecting spindle runout abnormality according to an embodiment of the present application, for performing diagnosis and early warning on a spindle of a CNC machine, as shown in fig. 5, the device includes:
the sensor 51 is arranged on the spindle and used for collecting a vibration data sample set of the spindle in a preset time period after multiple autorotations, the sensor in the embodiment may be a three-axis sensor or a six-axis sensor, and the sensor is arranged on the outer surface or the inner surface of the spindle to collect vibration data of the spindle in real time.
The processor 52 is coupled to the sensor 51, for example, connected to the sensor 51 by wire or wirelessly.
A memory 53 in which a plurality of program modules are stored; (the built prediction classification model, the updating module and the vibration data of the main shaft deflection collected by the sensor are stored in a supplementary mode, and the updating module is used for continuously updating the prediction classification model along with the continuous increase of historical data.)
The processor 52 is configured to load the plurality of program modules and execute the method for detecting spindle yaw abnormality according to the technical solution of the method embodiments shown in fig. 2 to fig. 4 in the description.
The sensor 51 is connected or communicatively connected to a processor 52 and a memory 53, and the processor 52 is connected or communicatively connected to the memory 53.
In one example of the present application, the processor 52 and the memory 53 may be provided in an electronic device.
The apparatus provided in the above-mentioned illustrated embodiment is used to implement the technical solution of the above-mentioned illustrated method embodiment, and the implementation principle and technical effect thereof may further refer to the related description in the method embodiment, which is not described herein again.
The apparatus provided in the above-described illustrated embodiment may be, for example: a chip or a chip module. The apparatus provided in the above-described embodiment is configured to implement the technical solution of the above-described method embodiment, and the implementation principle and technical effect of the apparatus may further refer to the relevant description in the method embodiment, which is not described herein again.
Each module/unit included in each device described in the above embodiments may be a software module/unit, or may also be a hardware module/unit, or may also be a part of a software module/unit, and a part of a hardware module/unit. For example, for each device applied to or integrated into a chip, each module/unit included in the device may be implemented by using hardware such as a circuit, or at least a part of the modules/units may be implemented by using a software program, where the software program runs on a processor integrated within the chip, and the remaining part of the modules/units may be implemented by using hardware such as a circuit; for each device applied to or integrated in the chip module, each module/unit included in the device may be implemented in a hardware manner such as a circuit, and different modules/units may be located in the same component (e.g., a chip, a circuit module, etc.) or different components of the chip module, or at least part of the modules/units may be implemented in a software program, where the software program runs on a processor integrated inside the chip module, and the rest of the modules/units may be implemented in a hardware manner such as a circuit; for each device applied to or integrated in the electronic terminal equipment, each module/unit included in the device may be implemented by using hardware such as a circuit, and different modules/units may be located in the same component (e.g., a chip, a circuit module, etc.) or different components in the electronic terminal equipment, or at least part of the modules/units may be implemented by using a software program running on a processor integrated in the electronic terminal equipment, and the rest (if any) part of the modules/units may be implemented by using hardware such as a circuit.
An embodiment of the present application provides a computer-readable storage medium, where the computer-readable storage medium stores computer instructions, and the computer instructions enable the computer to execute the method for detecting spindle yaw abnormality provided in the embodiments shown in fig. 2 to 4 in this specification. Computer-readable storage media may refer to non-volatile computer storage media.
The computer-readable storage medium described above may take any combination of one or more computer-readable media. The computer readable medium may be a computer readable signal medium or a computer readable storage medium. A computer readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium would include the following: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a Read Only Memory (ROM), an Erasable Programmable Read Only Memory (EPROM) or flash memory, an optical fiber, a portable compact disc read only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the context of this document, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.
Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, radio Frequency (RF), etc., or any suitable combination of the foregoing.
In the description of embodiments of the present application, reference to the description of "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present specification. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.
Further, in the description of the present specification, "a plurality" means at least two, for example, two, three, and the like, unless specifically defined otherwise.
Any process or method descriptions in flow charts or otherwise described herein may be understood as representing modules, segments, or portions of code which include one or more executable instructions for implementing steps of a custom logic function or process, and alternate implementations are included within the scope of the preferred embodiment of the present description in which functions may be executed out of order from that shown or discussed, including substantially concurrently or in reverse order, depending on the functionality involved, as would be understood by those reasonably skilled in the art of the embodiments of the present description.
It should be noted that the terminal referred to in the embodiments of the present application may include, but is not limited to, a Personal Computer (PC), a Personal Digital Assistant (PDA), a wireless handheld device, a tablet computer (tablet computer), a mobile phone, an MP3 player, an MP4 player, and the like.
The integrated unit implemented in the form of a software functional unit may be stored in a computer readable storage medium. The software functional unit is stored in a storage medium and includes several instructions to enable a computer device (which may be a personal computer, a server, or a network device) or a processor (processor) to execute some steps of the methods described in the embodiments of the present disclosure. And the aforementioned storage medium includes: a U disk, a removable hard disk, a Read Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, an optical disk, or the like.
The above description is only a preferred embodiment of the present disclosure, and should not be taken as limiting the present disclosure, and any modifications, equivalents, improvements, etc. made within the spirit and principle of the present disclosure should be included in the scope of the present disclosure.
Claims (10)
1. A method for detecting abnormal spindle deflection is used for diagnosing and early warning a spindle of a CNC machine station, and is characterized by comprising the following steps:
collecting vibration data sample sets of a plurality of main shafts of CNC machine tables of different types in a preset time period of multiple autorotation and the deflection radius of each main shaft;
extracting a time domain characteristic set in the vibration data sample set to form a training sample set, wherein the training sample set comprises a plurality of training sample units, and the training sample units comprise time domain characteristics corresponding to vibration data and corresponding deflection radiuses;
performing model training on the training sample set by using a gradient lifting tree algorithm to obtain a prediction classification model, wherein the model training comprises the following steps: inputting the training sample unit into a model determined based on a loss function and the gradient lifting tree algorithm, performing iteration of a preset turn to update model parameters, and determining the prediction classification model according to the model parameters;
collecting real-time vibration data of the main shaft in real time, and extracting corresponding time domain characteristics;
and loading the time domain characteristics corresponding to the real-time vibration data to the prediction classification model to obtain the deflection radius of the current main shaft, so as to predict the deflection level of the main shaft.
2. The detection method according to claim 1, wherein the step of inputting the training sample unit into a model determined based on a loss function and the gradient lifting tree algorithm and performing a preset number of iterations to update model parameters comprises:
constructing a decision tree for the time domain features based on the gradient lifting tree algorithm;
forming a learner f based on the loss function through the iteration of t-1 t-1 (x) Wherein T is an iteration round number, T is more than or equal to 2 and less than or equal to T, T is a preset round number, and x is a feature element in the time domain feature in the training sample unit;
after the t-th iteration, when the preset condition is met and the decision tree is established, the learner f t-1 (x) To form a learner f (x) and to update the model parameters.
3. The detection method according to claim 2, wherein after the t-th iteration, when a preset condition is met and the decision tree is established, the learner f performs the above-mentioned process t-1 (x) Foundation of (2)The step of updating to form the learner f (x) includes:
loss function pair learning device f based on current training sample unit to current t round number t-1 (x) Solving a first derivative and a second derivative, and substituting the first derivative and the second derivative into a loss function to obtain the value of the current loss function;
attempting to split the decision tree again based on the current node, wherein the decision tree comprises a plurality of leaf nodes, the default current score is 0, the current score is compared with a value corresponding to the current loss function, the leaf nodes with larger score are further split into subtrees, when the larger value is 0, the decision tree is completely built, and the learner f is provided with a function for splitting the decision tree again t-1 (x) Is updated to form the learner f (x).
4. The detection method according to claim 3, wherein the step of splitting the decision tree further comprises:
and randomly selecting s characteristic elements from the time domain characteristics in the current training sample unit, arranging the s characteristic elements according to the sequence from small to large, and placing the s characteristic elements into the left sub-tree and the right sub-tree of the decision tree to split the decision tree.
5. The method of claim 1, wherein the step of performing a predetermined number of iterations to update the model parameters further comprises: optimizing the preset turns using a stochastic search algorithm to optimize the model parameters.
6. The detection method of claim 1, wherein the time-domain feature comprises at least one of a peak factor, a pulse factor, a margin factor, a waveform factor, a kurtosis factor, and a skewness factor of the vibration data.
7. The method of claim 1, wherein the step of determining the predictive classification model based on the model parameters further comprises: performing K-fold cross validation on the training sample set, dividing the training sample set into K parts, selecting K-1 parts for training each time, and taking the rest parts as a test set to obtain the prediction classification model; and selecting K-1 parts again to be used as a training set, and using the rest parts as a testing set to optimize the prediction classification model.
8. The detection method according to claim 1, further comprising:
and when the deflection level of the current time based on the predicted deflection radius is lower than the deflection level of the main shaft before the current time, judging that the main shaft deflection is abnormal, and giving an early warning instruction to the CNC machine.
9. A detection device for abnormal spindle deflection is used for diagnosing and prewarning a spindle of a CNC machine table, and is characterized in that the prewarning device comprises:
the sensor is arranged on the main shaft and used for collecting a vibration data sample set of the main shaft in a preset time period of multiple autorotation;
a memory for storing a plurality of program modules;
a processor coupled to the sensor and the memory, wherein the processor is configured to load the plurality of program modules and execute the method for detecting spindle yaw abnormality according to any one of claims 1 to 8.
10. A computer readable storage medium having stored thereon at least one computer instruction, wherein the instruction is loaded by a processor and performs a method of spindle yaw anomaly detection according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211364856.8A CN115716217B (en) | 2022-11-02 | 2022-11-02 | Method and device for detecting spindle runout abnormality and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211364856.8A CN115716217B (en) | 2022-11-02 | 2022-11-02 | Method and device for detecting spindle runout abnormality and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115716217A true CN115716217A (en) | 2023-02-28 |
| CN115716217B CN115716217B (en) | 2023-09-01 |
Family
ID=85254607
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211364856.8A Active CN115716217B (en) | 2022-11-02 | 2022-11-02 | Method and device for detecting spindle runout abnormality and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115716217B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108984893A (en) * | 2018-07-09 | 2018-12-11 | 北京航空航天大学 | A kind of trend forecasting method based on gradient method for improving |
| WO2019001220A1 (en) * | 2017-06-26 | 2019-01-03 | 江南大学 | Decision tree based turning process energy consumption prediction system and method of numerically controlled lathe |
| CN112757053A (en) * | 2020-12-25 | 2021-05-07 | 清华大学 | Model fusion tool wear monitoring method and system based on power and vibration signals |
| US20210239577A1 (en) * | 2020-02-03 | 2021-08-05 | Airbus Operations (S.A.S.) | System for estimating the state of wear of a cutting tool during machining |
| CN114354184A (en) * | 2021-12-28 | 2022-04-15 | 哈尔滨工业大学 | Deep learning-based method and device for establishing health early warning model of main shaft of large-scale rotating equipment |
-
2022
- 2022-11-02 CN CN202211364856.8A patent/CN115716217B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019001220A1 (en) * | 2017-06-26 | 2019-01-03 | 江南大学 | Decision tree based turning process energy consumption prediction system and method of numerically controlled lathe |
| CN108984893A (en) * | 2018-07-09 | 2018-12-11 | 北京航空航天大学 | A kind of trend forecasting method based on gradient method for improving |
| US20210239577A1 (en) * | 2020-02-03 | 2021-08-05 | Airbus Operations (S.A.S.) | System for estimating the state of wear of a cutting tool during machining |
| CN112757053A (en) * | 2020-12-25 | 2021-05-07 | 清华大学 | Model fusion tool wear monitoring method and system based on power and vibration signals |
| CN114354184A (en) * | 2021-12-28 | 2022-04-15 | 哈尔滨工业大学 | Deep learning-based method and device for establishing health early warning model of main shaft of large-scale rotating equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115716217B (en) | 2023-09-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109297689B (en) | Large-scale hydraulic machinery intelligent diagnosis method introducing weight factors | |
| CN111459700B (en) | Equipment fault diagnosis method, diagnosis device, diagnosis equipment and storage medium | |
| CN104390657B (en) | A kind of Generator Unit Operating Parameters measurement sensor fault diagnosis method and system | |
| CN108280543B (en) | Working condition self-adaptive equipment health degree evaluation method based on classification regression mode | |
| CN108920863A (en) | A kind of method for building up of robot servo system energy consumption estimation model | |
| CN116448419A (en) | Zero sample bearing fault diagnosis method based on depth model high-dimensional parameter multi-target efficient optimization | |
| CN110795780A (en) | XGboost algorithm-based cable-stayed bridge finite element correction method | |
| CN114398049A (en) | Self-adaptive dynamic updating method for digital twin model of discrete manufacturing workshop | |
| CN112039073B (en) | Collaborative optimization method and system suitable for fault judgment of power distribution room equipment | |
| CN108961460B (en) | Fault prediction method and device based on sparse ESGP (Enterprise service gateway) and multi-objective optimization | |
| Qin et al. | Remaining useful life prediction for rotating machinery based on optimal degradation indicator | |
| Xiao et al. | Long-term predictive opportunistic replacement optimisation for a small multi-component system using partial condition monitoring data to date | |
| Lutska et al. | Forecasting the efficiency of the control system of the technological object on the basis of neural networks | |
| CN112763215B (en) | Multi-working-condition online fault diagnosis method based on modular federal deep learning | |
| CN104504265B (en) | A kind of method of servicing bridges monitoring information security evaluation | |
| CN117491503A (en) | Acoustic emission sensing-based laser powder feeding and material adding process monitoring and defect sensing method | |
| CN117035739A (en) | Work order path planning method and device for photovoltaic power station, electronic equipment and medium | |
| CN112128950A (en) | Machine room temperature and humidity prediction method and system based on multiple model comparisons | |
| CN117232809A (en) | Fan main shaft fault pre-diagnosis method based on DEMATEL-ANP-CRITIC combined weighting | |
| CN115716217B (en) | Method and device for detecting spindle runout abnormality and storage medium | |
| CN116720079A (en) | Wind driven generator fault mode identification method and system based on multi-feature fusion | |
| CN114580087B (en) | Method, device and system for predicting federal remaining service life of shipborne equipment | |
| US11339763B2 (en) | Method for windmill farm monitoring | |
| JP2020086786A (en) | Detection device and machine learning method | |
| CN115422821A (en) | Data processing method and device for rock mass parameter prediction |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |