Method and system for controlling a production plant to manufacture a product
A machine learning module is provided which is trained to generate from a design data record specifying a design variant of a product, a first performance signal quantifying a predictive performance of the design variant and a predictive uncertainty of the predictive performance. A variety of design data records each specifying a design variant of the product is generated. For a respective design data record, the following steps are performed: a first performance signal and a corresponding predictive uncertainty are generated, depending on the predictive uncertainty, a simulation yielding a second performance signal quantifying a simulated performance of the corresponding design variant is either run or skipped, and a performance value is derived from the second performance signal if the simulation is run or, otherwise, from the first performance signal. Depending on the derived performance values, a performance-optimizing design data record is determined and output to control the production plant.
1 . A computer-implemented method for controlling a production plant to manufacture a product, the method comprising:
a) providing a machine learning module which is trained to generate from a design data record specifying a design variant of the product:
a first performance signal quantifying a predictive performance of the design variant, and
a predictive uncertainty of the predictive performance;
b) generating a variety of design data records each specifying the design variant of the product;
c) for a respective design data record:
generating the first performance signal and a corresponding predictive uncertainty by the machine learning module,
depending on the predictive uncertainty being below a threshold value, skipping the simulation of the corresponding design variant, and depending on the predictive uncertainty exceeding the threshold value, running the simulation of the corresponding design variant, the simulation yielding a second performance signal quantifying a simulated performance of that design variant, and
deriving a performance value from the second performance signal if the simulation is run or, otherwise, from the first performance signal;
d) depending on the derived performance values, determining from the variety of design data records a performance-optimizing design data record; and
e) outputting the performance-optimizing design data record for controlling the production plant.
2 . The method according to claim 1 , wherein the machine learning module comprises or implements a surrogate model, a Bayesian machine learning model, a Bayesian neural network and/or a Gaussian process model.
3 . The method according to claim 1 , wherein the training of the machine learning module is continued by using the second performance signals as training data.
4 . The method according to claim 1 , wherein the predictive and/or simulated performance of a design variant depends on a fulfillment of one or more design objectives by the design variant and/or on a compliance of the design variant with one or more design constraints.
5 . The method according to claim 1 , wherein the predictive uncertainty is represented by a probability distribution, a discrete probability distribution, a statistical variance, a statistical standard deviation, a confidence interval, and/or an error interval.
6 . The method according to claim 1 , wherein for a decision whether to run or to skip a respective simulation, the corresponding first performance signal is taken into account.
7 . The method according to claim 1 , wherein for a decision whether to run or to skip a respective simulation, a deviation of a predictive performance quantified by the corresponding first performance signal from a predictive performance quantified by a previously generated first performance signal is taken into account.
8 . The method according to claim 6 , wherein an upper performance threshold is determined for a predictive performance quantified by the corresponding first performance signal and a corresponding predictive uncertainty,
a lower performance threshold is determined for a predictive performance quantified by a previously generated first performance signal and a corresponding predictive uncertainty, and
for the decision whether to run or to skip the respective simulation the upper performance threshold is compared with the lower performance threshold.
9 . The method according to claim 8 , wherein the upper performance threshold is ranked within one or more lower performance thresholds determined for one or more previously generated first performance signals, and
for the decision whether to run or to skip the respective simulation the rank is taken into account.
10 . A system for controlling a production plant to manufacture a product performing the method according to claim 1 .
11 . A computer program product, comprising a computer readable hardware storage device having computer readable program code stored therein, said program code executable by a processor of a computer system to implement a method according to claim 1 .
12 . The method according to claim 1 , further comprising: controlling the production plant using the performance-optimizing design data record to manufacture the product as specified by the performance-optimizing design data record.