Cable burning smell continuous detection method based on inspection robot

Abstract

The invention discloses a cable burning smell continuous detection method based on an inspection robot. The method comprises steps of burning different types of cables at high temperature, then detecting the gas generated by the high-temperature burning of the cables through a gas sensor module, collecting a corresponding smell data sample, forming a discrete time sequence signal, marking the odor data as burning odor of the cable in the odor environment, and training the odor data as a sample through a recurrent neural network, namely continuously collecting odor concentration data, forming a time sequence signal by the odor data according to a time sequence, and taking the time sequence signal as input of the recurrent neural network; then, a mathematical algorithm being combined to enable the recurrent neural network to be self-fitted to an information state capable of highlighting the change of the detected gas characteristics along with time, at the moment, after gas data are inputted, concentration characteristics being rapidly extracted, a classification decision being made, and gas detection and classification being achieved.

Description

technical field [0001] The invention belongs to the gas detection technology, in particular to a method for continuously detecting cable burnt smell based on an inspection robot. Background technique [0002] Cable burnt odor identification belongs to specific odor identification. At present, for odor data collection, most of them use sensitive membrane materials to absorb gas molecules to generate vibration, and what is sampled is a complex time series signal. This signal will be affected by various factors such as sensitive membrane material, gas type and concentration, and external environment (such as temperature and humidity). For the sampled data, manually designed features are usually used, combined with wavelet decomposition, support vector machine and other methods for identification; some improved algorithms independently design feature extraction and classification, and the process of feature extraction includes manual design features, wavelet decomposition and pr...

AI Technical Summary

Problems solved by technology

[0004] (1) Feature extraction is difficult: when there are too many types of odors, the classification and recognition results obtained by using traditional methods are often not ideal, that is, the features extracted by these common algorithms cannot truly represent the characteristics of material odor information

[0005] (2) Poor robustness of traditional methods and feature extraction: Traditional methods are mostly based on empirical rules, and cannot achieve good segmentation results when some gas components are similar, and the on-site interference environment will further aggravate the difficulty of feature extraction

For example, if the classification effect map is a five-point feature distribution map, it is based on static gas features. It cannot have a good extraction effect and feature description for continuously changing features, and if the changing gas is described by static features, its scope of application is relatively narrow. In the robot Problems cannot be found completely during inspection

[0006] (3) The feature description of ordinary deep learning methods is relatively static: while gas detection is continuous sampling in the air, which is approximately a continuous signal, the existing methods cannot describe the characteristics of gas changes from the time dimension, that is, continuous dynamic detection is not possible

Images