Introduction of intelligent evaluation technology for NC machining accuracy

In order to ensure the accuracy of NC machining, we conduct data processing on the three-dimensional acceleration sensor and other signals according to the signal time-domain analysis and frequency-domain analysis methods, extract the eigenvalues, and normalize the obtained eigenvalues, establish a neural network model, input the eigenvalues into the neural network for training, input the features to be tested after the training, and output the accuracy evaluation decision results from the neural network.

Intelligent evaluation system model

According to the hardware system, the machining accuracy evaluation model is established. The model is composed of different layers, including signal acquisition layer, signal output layer, signal transformation layer, signal conditioning layer, data acquisition layer, acquisition software, data storage, feature extraction and user layer.

The functions of each part are as follows:

(1) Signal acquisition layer: it is mainly used for each sensor to collect corresponding signals from the measuring points at the installed positions, and the signals output by the sensors are transmitted to the signal output layer.

(2) Signal input layer: the signal is transmitted to the discharge conditioning circuit of the CNC machine tool, and the signal output layer links the signal measuring point and the preprocessing circuit.

(3) Signal transformation layer: it can realize the transformation of signal form. Since the original signals output by each sensor include voltage signal, resistance signal and current signal, in order to facilitate data acquisition, these signals need to be transformed in the signal transformation layer and uniformly converted into voltage signals.

(4) Signal conditioning layer: mainly composed of signal conditioning instrument. Since the original signal is mixed with a large number of noise signals and the original signal is relatively weak, the signal conditioning layer mainly realizes amplification and filtering of the original signal.

(5) Data acquisition layer: mainly composed of data acquisition card to realize high-speed signal acquisition.

(6) Acquisition software: it is mainly used to realize automatic data acquisition, transmission, storage and other operations.

(7) Data storage: it is the basic basis for data processing. The stored data needs to be called in subsequent processing.

(8) Feature extraction: it is mainly used to extract relevant time-domain features and frequency-domain features from the processed signal for subsequent neural network training.

 (9) User layer: the neural network is mainly used to train and learn the extracted feature values and output the decision results.

Signal feature extraction

The feature selection value is to extract the feature information that can best reflect the change of machining accuracy from the original signal by using various digital signal analysis and processing methods and the relationship between the change of machining accuracy and the feature quantity. The original signal collected by the sensor contains a large number of noise signals. In order to effectively extract the characteristic value of the signal, wavelet packet is selected to extract the characteristic value.