| CPC G05B 23/024 (2013.01) [G05B 23/0221 (2013.01); G05B 23/0272 (2013.01); G05B 23/0281 (2013.01); G06F 18/214 (2023.01); G06N 3/08 (2013.01); G06F 18/24 (2023.01)] | 10 Claims |
|
1. A machine abnormality marking and abnormality prediction system connected with a host connecting to machines in a factory, comprising:
a parameter streaming unit connected with the machines and comprising a streaming server, a protocol server, a database server and a static server, wherein the streaming server is connected with each machine of the machines through a gateway device and comprises a shutdown module and maintenance module to transmit data of shutdown and maintenance of said each machine in operation through the protocol server; the database server is connected with the protocol server to periodically store data of said each machine in operation from the protocol server; and the static server is connected with the database server to count and average the data of said each machine in operation with a specific period;
an abnormality reporting unit comprising at least one handheld device and wirelessly connected with a streaming server of the parameter streaming unit, wherein the abnormality reporting unit transmits abnormality reasons and occurrence time caused by the shutdown of said each machine in operation to the shutdown module for storage, and transmits the abnormality reasons and occurrence time to the maintenance module for recording;
a prediction analysis unit comprising a microprocessor, a time sequence recorder and a neural network classifier, wherein the microprocessor is connected with the static server and the abnormality reporting unit to store the data of said each machine in operation analyzed and averaged with the specific period by the static server as historical data value, and store the abnormality reasons and occurrence time caused by the shutdown and maintenance of said each machine in operation reported by the abnormality reporting unit as instant data value which serves as a prediction value for training the neural network classifier; and the time sequence recorder is connected with the microprocessor to extract one-dimensional vectors and combine the one-dimensional vectors in series with specific second, minute and hour as time sequence; and
the neural network classifier connected with the time sequence recorder and comprising a first trigger body and second trigger body to continuously use a minimal difference to predict a cross entropy loss in the instant data value and the historical data value in order to optimize a neural network module and accordingly predict states of machines, so as to generate abnormality warnings, for factory workers to arrange maintenance or adjust machines of factory production lines in advance to avoid occasional shutdown and reduce a factory loss.
|