LSTM and CNN based tool wear state predicating method and device

Abstract

The invention relates to an LSTM and CNN based tool wear state predicating method and device. The method comprises the following steps: acquiring original data in a machine tool processing process, wherein the original data comprise machine tool vibration data, tool cutting force data and high-frequency stress wave data; inputting the original data into an LSTM network, and extracting time sequence characteristics of the original data; inputting the original data with time sequence characteristics extracted into the CNN network, and extracting multi-dimensional characteristics comprising the time sequence characteristics from the original data; and calculating to obtain a tool wear predicated value based on the multi-dimensional characteristics comprising the time sequence characteristicsand a set mapping relationship. The invention further discloses an LSTM and CNN based tool wear state predicating device. According to the LSTM and CNN based tool wear state predicating method and device disclosed by the invention, the extraction effect on the data multi-dimensional characteristics and the time sequence characteristics can be integrated, so that the tool wear state predicating effect can be improved.

Description

technical field [0001] The invention relates to the technical field of CNC machine tool tool identification, in particular to a tool wear state prediction method and device based on LSTM and CNN. Background technique [0002] CNC machine tool is the abbreviation of Computer numerical control machine tools, which is an automatic machine tool equipped with a program control system. CNC machine tools generally include the machine tool body, machine table, machine tool spindle and tools; and the prediction of tool wear status is very important, because excessive tool wear will increase the consumption of production resources and affect the quality of the workpiece. [0003] The existing tool wear state prediction method is to collect data reflecting tool wear through a variety of sensors, and analyze the collected data in time domain, frequency domain and time-frequency domain to extract features related to tool wear, such as mean Square root error value, variance, skewness, sp...

AI Technical Summary

Problems solved by technology

[0006] However, in the actual prediction process, the existing tool wear state prediction method still has the following problems: it first extracts the multi-dimensional features of the original data through the CNN network, and then extracts the time series features of the original data through the LSTM network, so that when extracting multiple When dimension features are used, the time series features of the original data have been destroyed, resulting in the loss of the sequence feature information of the original data, making it difficult to take into account the extraction effects of multi-dimensional features and time series features of the original data, resulting in poor prediction of tool wear status

Images