Detecting Eggshell Cracks by Acoustic Impulse Response and Artificial Neural Networks

Abstract

An experimental system was developed for detecting cracks in eggshells based onacoustic impulse response and artificial neural networks (ANNs). The eggs were lightly impactedeight times individually at equal distance around the equator ring to record the acoustic response ofegg surface. The frequency spectra of 4496 acoustic patterns obtained from 300 intact, 52 naturallycracked and 210 artificially cracked eggs were analyzed for training the backpropagation, Kohonenand probabilistic ANNs. Hand candling was used for examining the eggs to verify the presence ofsurface cracks prior to acoustic testing. Intact eggs produced sound signals mainly exhibiting a singledominant peak in the frequency range of 2500-7000 Hz with signal duration of about 20 ms. Thecracked eggs showed frequency spectra in relatively wider frequency range of 550-9000 Hz andshorter signal duration of about 15 ms. In the training of ANNs, 188 neurons were used in the inputlayer representing signal amplitudes in the frequency range of 597-9173 Hz at 46 Hz interval, andtwo neurons in the output layer representing either a cracked or intact egg. A backpropagation ANNwith three hidden slabs showed best performance with 98.7% crack detection accuracy and 10%false rejects.