Neural network-based acoustic glass defect detection method

Abstract

The invention discloses a neural network-based acoustic glass defect detection method which comprises the following steps: acquiring a knocking signal of a glass sample through a pickup in an actual production environment; preprocessing the knocking signal; performing feature extraction on the pure knocking signal; setting initial parameters of a BP neutral network: taking an extracted feature as the input of the neutral network, setting the number of nodes on an input layer of the BP neutral network to be 7, setting the number of nodes on a hidden layer of the BP neutral network to be 15, setting the number of nodes on an output layer of the BP neutral network to be 2, and setting an output result to be (0, 1) which indicates that the glass sample has a defect, and to be (1, 0) which indicates that the glass sample is defect-free; training the BP neutral network, setting the learning rate of the BP neutral network to be 0.1, setting a target square error value to be 0.1, training the BP neutral network through a LeVenberg-Marquardt algorithm, and stopping training if an error of the neutral network is less than the set target square error value. A signal feature extracted through the method is high in distinction degree, and can complete a glass defect detection task more accurately and efficiently.

Description

technical field [0001] The invention relates to a glass defect detection method. In particular, it relates to a neural network-based acoustic glass defect detection method. Background technique [0002] Glass is one of the most common materials, widely used in various industrial fields, but also one of the most fragile and easily damaged materials. During the production and transportation of glass products, defects such as holes, cracks, and trachoma will be caused. If the defective products cannot be removed before sale, the qualified rate of the products will be reduced and certain economic losses will be brought. [0003] At present, the non-destructive testing methods for glass product defects mainly include manual methods, computer vision testing methods, ultrasonic testing methods, vibration testing methods, etc. The manual method relies on experienced workers to inspect the product. This method has the following disadvantages: (1) The misjudgment rate is high. Du...

AI Technical Summary

Problems solved by technology

This method has the following disadvantages: (1) The misjudgment rate is high

Due to the high labor intensity of the workers, the human eyes are prone to fatigue when observing for a long time, which will directly affect the accuracy of classification

(2) Judgment standards are not uniform. During manual inspection, for the case of small defects, whether it is qualified or not is judged subjectively by the inspector. It is difficult to unify the judgment standards of different people or even the same person under different conditions.

[0005] The current existing methods mainly use the correlation of the signal in the channel on the time axis to collect signals, but due to the instability of the sound signal, the correlation on the time axis is not good, resulting in unsatisfactory results after sparse processing.

Images