Development and evaluation of an expert system for egg sorting

Neural networks were previously developed for the detection of blood spot, crack and dirt stain defects in eggs. In the present study, an expert system was developed to sort eggs into pre-defined categories based on the intended use of the eggs. The expert system used the output values of the neural networks and its knowledge-base to make decisions on how to sort the eggs. The sorting decisions of the expert system were also influenced by several variable thresholds which could be set by the user. To demonstrate and test the expert system under varying threshold settings, a discrete computer simulation of the egg inspection process was performed. In the simulation, a sample of 10000 egg entities was generated. The number of egg entities with defects corresponded to experimentally observed defect frequencies. The inspection of the egg entities by the neural networks was simulated by assigning each egg entity a random sample from distributions of observed output values of the trained neural networks. The expert system was used to sort the sample of egg entities. In the simulation study, the expert system sorted the egg entities into five categories. Lower threshold values resulted in more egg entities being rewashed, inspected or rejected for bakery or animal use, whereas higher threshold settings caused more egg entities to be sorted into the 'grade A' category. In the simulation study, the expert system assigned a large number of egg entities to be rewashed. This indicated a need for improvement of the dirt stain detection neural network. The simulation results indicated a reduction in the number of eggs requiring human inspection and demonstrated a potential for reducing the work load on human graders. (C) 1998 Elsevier Science B.V. All rights reserved.