Damage prediction using neural networks

The ability to predict damage during package distribution becomes critical to the success of many industries. The problem is complicated by the multip le variables involved and uncertain interactions between them. The lack of scientific and efficient tools has lead to the common practice of over-pack aging, which results in significant cost increase and solid waste. A neural network model capable of predicting damage rate of packaged products is de veloped in this study. It has a simple structure and is trained with experi mental data. The network is capable of predicting damage rate given the inp uts such as hazard type, loading level, cushioning and package material. Di fferent techniques are used to speed up the learning process and improve th e model performance. Test results show the validity and consistency of the model developed. Significance: This study developed a neural network that can be used to pre dict damage rate of products in package distribution and provide more objec tive and accurate results in an efficient way. This facilitates decisionmak ing in the design and operation of distribution packaging. The modeling and improvement analysis also provide some valuable insight on engineering app lication of neural network, which can be used to enhance the understanding and develop more powerful models for related problems.