Press machine and method for monitoring abnormality of press machine
A press machine includes: a learning-model generating unit that uses one data from among data collected from sensors, as an objective variable, and uses data other than the one data as an explanatory variable to perform machine learning to generate a learning model for the one data, the generation being performed for all the data; a predicted-value calculating unit that inputs an actually measured value of data other than one data from among the data collected from the sensors, into the learning model for the one data to calculate a predicted value of the one data, the calculation being performed for all the data; a degree-of-abnormality calculating unit that calculates a degree of abnormality based on a difference between an actually measured value and a predicted value of the data; and a degree-of-abnormality outputting unit that outputs the calculated degree of abnormality.
1. A press machine comprising:
a plurality of sensors collecting n pieces of correlated data;
a learning-model generating unit configured to, for each of the n pieces of correlated data, use one piece of correlated data as an objective variable, and use all other n−1 pieces of correlated data as an explanatory variable to perform machine learning to generate a learning model for the one piece of correlated data;
a predicted-value calculating unit configured to, for each of the n pieces of correlated data, calculate a predicted value of the one piece of correlated data by inputting actually measured values of all other n−1 pieces of correlated data into the learning model for the one piece of correlated data;
a degree-of-abnormality calculating unit configured to calculate a degree of abnormality as a weighted sum of differences between the actually measured values and the predicted values of each of the n pieces of correlated data; and
a degree-of-abnormality outputting unit configured to output the calculated degree of abnormality.
2. The press machine according to claim 1 , wherein
the press machine is a servo press, and
the n pieces of correlated data include data on a press load and data on output current of a servo amplifier.
3. The press machine according to claim 2 , wherein
the n pieces of correlated data include at least one of data on input current to a servo power supply, data on voltage across PN, data on a temperature of the servo power supply, data on a temperature of a servo amplifier, and data on a temperature of a capacitor.
4. The press machine according to claim 2 , wherein the n pieces of correlated data include at least one of data on input current to a servo power supply, data on voltage across PN, data on a temperature of the servo power supply, data on a temperature of a servo amplifier, and data on a temperature of a capacitor, data on an ambient temperature, and data on a temperature of a lubricating oil.
5. The press machine according to claim 1 , wherein
the press machine is a mechanical press; and
the n pieces of correlated data include data on a press load and data on an output current of an inverter.
6. The press machine according to claim 5 , wherein the n pieces of correlated data include at least one of data on an ambient temperature, data on a temperature of lubricating oil, and data on a temperature of a clutch and brake.
7. A method for monitoring an abnormality of a press machine, comprising:
collecting n pieces of correlated data by a plurality of sensors;
for each of the n pieces of correlated data, using one piece of correlated data as an objective variable, and using all other n−1 pieces of correlated data as an explanatory variable to perform machine learning to generate a learning model for the one piece of correlated data;
for each of the n pieces of correlated data, calculating a predicted value of the one piece of correlated data by inputting actually measured values of all other n−1 pieces of correlated data into the learning model for the one piece of correlated data;
calculating a degree of abnormality as a weighted sum of differences between the actually measured values and the predicted values of each of the n pieces of correlated data; and
outputting the calculated degree of abnormality.