Context:

There is a growing need to monitor and predict critical variables in production processes, from the statistical process control approach it has been assumed from the use of control charts for individual measurements, for that reason this article presents the results of the design of a long short term memory (LSTM) recurrent neural network to predict the average value of the variable temperature in individual measurements and thus evaluate the ability of the network to obtain values similar to the EWMA weighted moving average calculations for individual measurements. Being this

Methodology:

1768 records of individual temperature measurements made by a sensor were obtained, in the data set called: Gas sensors for home activity monitoring data set. Temperature data was plotted on an EWMA exponential weighted moving average control chart to obtain process mean values and identify that the process was within statistical control. Subsequently, an LSTM neural network was trained on a training sample of 1184 data with a Backpropagation algorithm that allowed obtaining values similar to EWMA, which were validated in a test sample of 584 temperature data.

Results:

The design of a neural network with a unit in the input gate, 4 units in the forgetting gate and 1 unit in the output gate was obtained, trained with the Backpropagation algorithm, it allowed to calculate values very close to those represented in the control chart. EWMA, with an MSE of 1.1405e-04.

Conclusions:

LSTM neural networks are a good alternative for calculating EWMA values, when statistical control of a process that generates a large amount of data obtained from measurements is required and there is no software to process them.

Financing:

Fundación universitaria Los Libertadores