Research On Micro-drilling Online Monitoring Based On Double Neural Network

The development of high-tech products continuously increases the number of microholes and impels more and more wide application of micro holes. Processing quality ofmicro holes, example surface quality, dimensional accuracy, positional accuracy and workingefficiency are taken more and more high requests. Micro holes processing faces enormouschallenge. Drilling process by twist drills, as a traditional holes processing method, ispopular with the major businesses, because of simple equipments, convenient operation,reasonable price. And it has many advantages such as high processing efficiency, goodsurface quality, high precision and processing of no restriction from conductive properties ofmaterials. Micro drilling becomes the method with the lowest production cost and the mosteconomical and practical among many micro holes processing methods.The structural characteristics of small diameters, large length-diameter ratio of microdrills determines some fatal weaknesses, including low strength, poor rigidity, easy breakagedifficultly drilling, drilling offsetting, chip plugging, and so on. These usually cause microdrills breakage or damage, bring adverse effects in practice. Micro drill is easy to break anddifficult to predict, therefore in the drilling process, in order to prevent their breakage ordamage and improve their life, many experts and scholars have been committed to studymicro drilling. They hope to find some better methods to solve the outstanding problems onmicro drilling. So the focus of this paper are how to safeguard micro drills, give full play tothe function of micro drills, improve micro drilling capacity, productivity and toolsutilization rate.Force signals are the necessary conditions to represent continuously cutting and thenecessary factors to character working state of micro drills. Real-time monitoring forcesignals can realize on-line monitoring the state of micro drills, so that effectively prevent andavoid their breakage.Because drilling force signals showed two changes under normal working and microdrills breakage, the drilling force signals are summed up two kinds of modes, one isprogressive type wear broken mode, another is mutant abnormal failure mode. Taking theseas the premise, the double neural network models are constructed, some characteristics of thesignal amplitudes of drilling thrust and torque signals and their first time derivative composes the input layer vectors of the neural network. The networks are trained by usingthe experimental data, two groups of monitoring threshold are obtained and used to on-linemonitor micro drilling.The paper gives full play to LabVIEW software advantages, combines the G-languageprogramming function and experimental device, a kind of micro drilling force signals on-linemonitoring system is designed based on VI technology. The system is composed of severalparts including train sensors, the charge amplifier, the A/D PCI, the computer softwaremonitoring system, the microprocessor control unit, the power amplifier, the drive power, thestep motor, the drive chain and the precision drilling machine. The system working steps areacquisition of signals, real-time displaying and historical data reproduction, data handling,signal analysis, neural network decision, servo control to the step motor and alarm backcutting tools.Drilling force monitoring experiments are done by using this system. Specific process isas follows, the first step is to comparing the real-time drilling force signals and thepredetermined values in order to choose the neural network model, second is that thereal-time data are input into the trained network model and train or calculate, third is tocontrast the network output to the given monitoring threshold, micro-drills will continue todrill while the output is less than the threshold, or else micro-drills will been alerted to drawback while the output is greater than or equal the threshold. The results showed that selectingappropriate monitoring threshold could effectively avoid micro-drills breakage. Somonitoring micro-drilling using the double neural network is feasible and available.In this paper, the main innovations are as follows:(1) A kind of experiment system is built to collect the drilling thrust and torque signalsby sensors. A lot of experimental results show that the more drilling holes, the more seriousmicro drills wear, the greater the drilling forces are. They also show that the amplitudes ofdrilling forces are sensitive to micro drills’ wear, so drilling forces may represent the wearingstate and cutting state of micro drills, drilling forces are usual as the character parameters. Inthe experiments, the two changes under normal working and micro drills breakage, thedrilling force signals are summed up two kinds of modes, one is progressive type wearbroken mode, another one is mutant abnormal failure mode. (2) The double neural network model, including the progressive type and the mutantabnormal failure mode, is set up to on-line monitoring micro drilling. The double neuralnetwork is trained by lots of experimental samples, the two stable networks are obtained. Aeffective method is get to intelligent on-line monitoring micro drilling.(3) A kind of micro drilling force signals on-line monitoring system is designed basedon the combination of the LabVIEW software and lots of hardware. Some contrast tests aredesigned based on between the single neural network and the double neural network, resultsshow that the effect on the double neural network is better than its effect on the single neuralnetwork during micro drilling on-line monitoring.