CN107553219A

CN107553219A - A kind of Tool Wear Monitoring method based on multiple types sensor composite signal

Info

Publication number: CN107553219A
Application number: CN201710990870.1A
Authority: CN
Inventors: 单春雷; 聂鹏; 李正强; 杨新岩
Original assignee: SHENYANG BAIXIANG MECHANICAL PROCESSING CO Ltd
Current assignee: SHENYANG BAIXIANG MECHANICAL PROCESSING CO Ltd
Priority date: 2017-10-23
Filing date: 2017-10-23
Publication date: 2018-01-09
Also published as: CN109318056A

Abstract

The present invention relates to a kind of Tool Wear Monitoring method based on multiple types sensor composite signal.The present invention is using the related signal message of acoustic emission sensor and power sensor collection lathe tool wear, the defects of by the method for two kinds of signal acquisitions single signal being avoided own in itself.Using two kinds of information of coupling of cloud model algorithm science, and the characteristic factor for reflecting tool abrasion in signal can be extracted, model is established using sparse Bayesian method and then predicts tool abrasion, data are modeled using the recognition methods based on SBL, the width parameter of SBL model kernel functions is optimized using Bayesian matching tracing algorithm, the Accurate Prediction of tool abrasion is realized, improves the efficiency and accuracy of Tool Wear Monitoring.

Description

A kind of Tool Wear Monitoring method based on multiple types sensor composite signal

Technical field

The present invention relates to a kind of Tool Wear Monitoring method based on multiple types sensor composite signal, belong to tool wear Detection field.

Background technology

One of the core of smart machine as wisdom factory, to the Urine scent of running status, self-teaching and self dimension Shield ability is its key character.According to statistics, tool changing and the 20% of operation hours is accounted for knife in process.In addition, knife The abrasion of tool and the damaged personal safety to crudy, processing efficiency, lathe life-span even operating personnel have a major impact.Cause This, accurately and efficiently cutter running status Urine scent and automatic early-warning are significant to the level of intelligence for improving lathe, can Effectively cost-effective, raising efficiency

Because the Tool Wear Process in high-speed milling is complicated, model parameter is excessive and is difficult to predict tool wear how Predict that tool wear turns into the focus of researcher by more efficient approach.The prediction of domestic and foreign scholars Cutter wear is done Substantial amounts of research, and achieve many progress.

Sound emission (Acoustic emission, AE) technology as advanced detection means to various metal materials and its Internal microcrack is very sensitive, can find the early changes of material, therefore is widely used in various plant equipment detections.But due to every Fault message contained by individual sound emission scatterplot is different, therefore also different to assessment equipment status information contribution degree；And each sound emission dissipates The characteristic parameter physical meaning of point extraction is different, also different to Fault-Sensitive degree.Some characteristic parameters can be sent out in failure early stage Raw mutation, and some characteristic parameter variation tendencies are relatively gentle, can not provide early warning for equipment failure state.

Another key problem of tool condition monitoring is on the basis of signal Analysis feature, builds effective algorithm and carries out The prediction of tool abrasion.Algorithms most in use includes artificial neural network and SVMs etc..The pre- measuring and calculating of artificial neural network Method, model is excessively complicated, needs substantial amounts of experiment sample, and calculating convergence difficulties SVMs can be realized in small sample The prediction of bottom tool wear extent, but easily study phenomenon occurred, model it is openness limited, and the general of prediction result can not be provided Rate information.

The content of the invention

In order to solve above-mentioned technical problem, the present invention provides a kind of knife based on multiple types sensor composite signal Has wear monitoring method.

The purpose of the present invention is achieved through the following technical solutions：A kind of knife based on multiple types sensor composite signal Has wear monitoring method, it is characterised in that comprise the following steps：

(1) data acquisition：Using the acoustic emission signal of acoustic emission sensor collection lathe, power sensor harvester is used Bed working power signal, while taken pictures after the processing period every time to cutter using microscope, and after measuring cutter Knife face attrition value, tool wear data are obtained for comparing；

(2) feature extraction：Denoising is carried out to the signal of collection using cloud algorithm, filters out interference band to extracting feature The influence of parameter, feature extraction then is carried out to data, the correlation of each feature and tool abrasion is analyzed and chooses correlation Strong feature；

(3) classify and predict after building model and optimization：Data characteristics by post processing is ground with cutter is measured microscopically Damage amount data form sample group, data are modeled using the recognition methods based on SBL, using Bayesian matching tracing algorithm The width parameter of SBL model kernel functions is optimized, realizes the Accurate Prediction of tool abrasion.

Beneficial effects of the present invention：The present invention is using acoustic emission sensor and power sensor collection lathe tool wear phase The signal message of pass, by the method for two kinds of signal acquisitions can avoid single signal in itself have by oneself the defects of.Using cloud model Two kinds of information of coupling of algorithm science, and the characteristic factor for reflecting tool abrasion in signal can be extracted, use sparse pattra leaves This method establishes model and then predicts tool abrasion, realizes the monitoring of Cutter wear, improves the effect of Tool Wear Monitoring Rate and accuracy.

Brief description of the drawings

Fig. 1 is the general construction block diagram of the invention.

Fig. 2 is the flow chart of Test Data Collecting of the present invention.

Embodiment

Step 1 data acquisition：

Data acquisition is carried out with U.S. PAC multiple channel acousto transmitting data collecting system, is passed sound emission by magnet base Sensor is fixed on testing stand girff, and first installation 1 cutter, gathers 10s acoustic emission signal and power signal in girff；More Tool changing has, and gathers remaining successively by same steps 8 the acoustic emission signal and power signal of cutter different periods.

In order to preferably study the forecasting problem in different workpieces processing conditions bottom tool state of wear and abrasion magnitude relation, Such as by 3 kinds of cutting parameters (cutting speed, the amount of feeding and back engagement of the cutting edge) global combinatorial, it will produce multigroup machining condition, cause Tested number is excessive.Therefore orthogonal experiment is used, under minimum test number (TN), scientifically arranges multigroup cutting parameter combination examination Test.

Step 2 feature extraction：

Denoising is carried out to the signal of collection using cloud algorithm, filters out influence of the interference band to extraction characteristic parameter. Then feature extraction is carried out to data, analyzes the correlation of each feature and tool abrasion and choose the strong feature of correlation.

Entropy reflects the degree of uncertainty of qualitativing concept corresponding to different wear stages, shows as signal in the abrasion Stage corresponds to the tolerance interval size of cloud concept.The entropy En of each cloud concept is calculated based on the reverse cloud algorithm without degree of certainty.

Choose the different state of wear reconstruction signal s ' (t) of gained under 5 groups of machining conditions and calculate discovery, increase entropy with wear extent Value En is in downward trend after first increasing.Abrasion initial stage, the entropy of clock signal is smaller, illustrates that the cloud concept of clock signal is covered The scope of lid is smaller, because cutter is very fast in abrasion abrasion early stage, quickly into mid-term wear stage；Hereafter, cutter is ground Damage progresses into mid-term stage, and the entropy of clock signal gradually increases, and illustrates that the scope that concept is covered becomes wide, this is due to this Stage tool wear is slower, and tool wear will enter one and smoothly normally cut the stage；Abrasion continues to aggravate, clock signal Entropy diminishes again, and cloud concept institute coverage reduces, because during into later stage wear stage, crash rate significantly raises, knife Tool abrasion is accelerated.

Step 3 feature post-processes：

In order to strengthen the susceptibility of data characteristics Cutter wear amount, data characteristics obtained above is post-processed, Including isotonic regression and exponential smoothing so that the feature after processing can better adapt to tool abrasion forecast model.

Step 4 builds classification and prediction after model and optimization：

Management loading be Tipping proposed on the basis of SVMs for the machine that returns and classify Learning method, SBL employ Bayesian inference method, and model has good openness, can avoid study phenomenon, and have simultaneously Probabilistic forecasting ability.

Experimental result and analysis

Characteristic after processing is divided into 2 groups, wherein 65% is training group, remaining is the mistake of the test group experiments Journey and parameter are as follows：

The study of sparse machine is the powerful for the model for obtaining the high dimensional data comprising bulk information, and its calculation cost compared with It is low.We do such hypothesis in experiment：If a connection weight is zero, then we are interpreted as it to be eliminated.

Experiment uses the Bayesian learning method of non-negative least square regular parameter.This method can be from noisy data Accurately obtain sparse solution.The length that the signal of collection is used in experiment is respectively 128, and degree of rarefication K=10 is that signal has individual position The coefficient put is not zero, and the dimension of observing matrix is 64*128 after feature is submitted, observation noise variance 0.05, and initiation parameter is 0.1。

With set according to being trained to the forecast model, with the degree of accuracy of test group data verification model, its is accurate Degree is more than 92%.

Claims

A kind of 1. Tool Wear Monitoring method based on multiple types sensor composite signal, it is characterised in that comprise the following steps：

(1) data acquisition：Using the acoustic emission signal of acoustic emission sensor collection lathe, added using power sensor collection lathe Work power signal, while cutter is taken pictures using microscope after the processing period every time, and measure knife face after cutter Attrition value, tool wear data are obtained for comparing；

(2) feature extraction：Denoising is carried out to the signal of collection using cloud algorithm, filters out interference band to extracting characteristic parameter Influence, then feature extraction is carried out to data, analyzes the correlation of each feature and tool abrasion and to choose correlation strong Feature；

(3) classify and predict after building model and optimization：Data characteristics by post processing and it is measured microscopically tool abrasion Data form sample group, data are modeled using the recognition methods based on SBL, using Bayesian matching tracing algorithm pair The width parameter of SBL model kernel functions optimizes, and realizes the Accurate Prediction of tool abrasion.