Determination device, deterioration determination system, work support device, deterioration determination method, and computer program

A deterioration determination method by a determination device comprising a storage and a processor, the deterioration determination method comprising:

• acquiring, by the processor, training data in which determination information is associated with a degree of deterioration or a degree of guarantee of a lead-acid battery, the determination information including at least one of appearance information or image information of the lead-acid battery;
• training, by the processor, a multilayer neural network by:
• inputting, in an input layer of the multilayer neural network, the determination information in the training data;
• outputting, from an output layer of the multilayer neural network, a determination result of a degree of deterioration or a degree of guarantee;
• comparing the determination result output from the output layer with a correct value which is information labeled with the determination information in the training data;
• and
• optimizing a parameter used for a calculation processing in an intermediate layer of the multilayer neural network so that an output value from the output layer approaches the correct value using an error back propagation method, the parameter being at least one of a coupling coefficient or a coefficient of an activation function;
• storing, by the processor, the trained multilayer neural network in the storage;
• receiving, by the processor, the determination information from a terminal connected to the determination device via a network;
• inputting, by the processor, the received determination information to the trained multilayer neural network;
• acquiring, by the processor, the degree of deterioration or the degree of guarantee of the lead-acid battery from the trained multilayer neural network;
• and
• retraining the trained multilayer neural network by:
• comparing the degree of deterioration or the degree of guarantee determined using the trained multilayer neural network and the degree of deterioration or the degree of guarantee obtained by actual measurement;
• if the degree of deterioration or the degree of guarantee determined using the trained multilayer neural network matches the degree of deterioration or the degree of guarantee obtained by actual measurement, inputting pieces of training data in which the determination result is associated with the degree of deterioration or the degree of guarantee determined using the trained multilayer neural network in the trained multilayer neural network;
• and
• if the degree of deterioration or the degree of guarantee determined using the trained multilayer neural network does not match the degree of deterioration or the degree of guarantee obtained by actual measurement, inputting training data in which the determination result is associated with the degree of deterioration or the degree of guarantee obtained by actual measurement in the trained multilayer neural network
• wherein the degree of deterioration includes information indicating that a deterioration is caused by a user's fault,
• the degree of guarantee includes:
• a first degree indicating that the lead-acid battery is returned to a user as a normal product;
• a second degree indicating that although the lead-acid battery is normal, the lead-acid battery is returned after performing auxiliary charge;
• a third degree indicating that the lead-acid battery is exchanged as an abnormal product;
• and
• a fourth degree indicating that although the lead-acid battery is an abnormal product, an abnormality has occurred due to an action of the user, and replacement is performed for a fee, and
• the multilayer neural network is configured to input the image information to a plurality of convolution layers and pooling layers to compress the image information, combine data of the image information of the pooling layers with other determination information, and then further input the image information to the plurality of convolution layers and pooling layers to compress the combined data of the image information and other determination information, and output the degree of deterioration or the degree of guarantee.