Controlling operation of machine tools using artificial intelligence

A computer implemented method comprising:

• using Deep Learning techniques to determine, for a factory comprising multiple machine tool workstations, a candidate sequence of current parts to be processed by the multiple machine tool workstations, comprising:
• training a neural network to process a data input representing multiple parts to be processed by the multiple machine tool workstations and generate a data output representing a candidate sequence of parts that, when processed by the multiple machine tool workstations, improves a value of a predefined performance metric for the multiple machine tool workstations, wherein the training of the neural network is performed using training examples that are determined offline using a branch and bound method;
• receiving data that represents the current parts to be processed by the machine tool workstations in the factory;
• processing the received data using the trained neural network to obtain the candidate sequence of the current parts to be processed;
• and
• verifying whether the determined candidate sequence of current parts to be processed will be delivered on time using Little's Law.

A system for comprising:

• one or more processing devices;
• one or more non-transitory machine-readable storage devices storing instructions that are executable by the one or more processing devices to cause performance of operations comprising:
• assigning machine tool workstations in a factory to multiple pull system groups, wherein each pull system group includes similar machine tool workstations;
• using Deep Learning techniques to determine, for the each pull system group of the multiple pull system groups, candidate sequences of current parts to be processed by the machine tool workstations assigned to the each pull system group, comprising:
• training, for the each pull system group of the multiple pull system groups, a respective neural network to process a data input representing multiple parts to be processed by the machine tool workstations assigned to the each pull system group and generate a data output representing a candidate sequence of parts that, when processed by the each pull system group, improves a value of a predefined performance metric for the machine tool workstations assigned to the each pull system group, wherein the training of each neural network is performed using respective training examples that are determined offline using a branch and bound method;
• receiving data that represents the current parts to be processed by the machine tool workstations in the factory;
• processing subsets of the received data using corresponding trained neural networks to obtain the candidate sequences of current parts to be processed;
• and
• verifying whether the determined candidate sequences of current parts to be processed will be delivered on time using Little's Law.

A system for comprising:

• one or more processing devices;
• one or more non-transitory machine-readable storage devices storing instructions that are executable by the one or more processing devices to cause performance of operations comprising:
• using Deep Learning techniques to determine, for a factory comprising multiple machine tool workstations, a candidate sequence of current parts to be processed by the multiple machine tool workstations, comprising:
• training a neural network to process a data input representing multiple parts to be processed by the multiple machine tool workstations and generate a data output representing a candidate sequence of parts that, when processed by the multiple machine tool workstations, improves a value of a predefined performance metric for the multiple machine tool workstations, wherein the training of the neural network is performed using training examples that are determined offline using a branch and bound method;
• receiving data that represents the current parts to be processed by the machine tool workstations in the factory;
• processing the received data using the trained neural network to obtain the candidate sequence of the current parts to be processed;
• and
• verifying whether the determined candidate sequence of current parts to be processed will be delivered on time using Little's Law.

A non-transitory machine-readable storage device storing instructions that are executable by a processing device to cause performance of operations comprising:

• assigning machine tool workstations in a factory to multiple pull system groups, wherein each pull system group includes similar machine tool workstations;
• using Deep Learning techniques to determine, for the each pull system group of the multiple pull system groups, candidate sequences of current parts to be processed by the machine tool workstations assigned to the each pull system group, comprising:
• training, for the each pull system group of the multiple pull system groups, a respective neural network to process a data input representing multiple parts to be processed by the machine tool workstations assigned to the each pull system group and generate a data output representing a candidate sequence of parts that, when processed by the each pull system group, improves a value of a predefined performance metric for the machine tool workstations assigned to the each pull system group, wherein the training of each neural network is performed using respective training examples that are determined offline using a branch and bound method;
• receiving data that represents the current parts to be processed by the machine tool workstations in the factory;
• processing subsets of the received data using corresponding trained neural networks to obtain the candidate sequences of current parts to be processed;
• and
• verifying whether the determined candidate sequences of current parts to be processed will be delivered on time using Little's Law.

A non-transitory machine-readable storage device storing instructions that are executable by a processing device to cause performance of operations comprising:

• using Deep Learning techniques to determine, for a factory comprising multiple machine tool workstations, a candidate sequence of current parts to be processed by the multiple machine tool workstations, comprising:
• training a neural network to process a data input representing multiple parts to be processed by the multiple machine tool workstations and generate a data output representing a candidate sequence of parts that, when processed by the multiple machine tool workstations, improves a value of a predefined performance metric for the multiple machine tool workstations, wherein the training of the neural network is performed using training examples that are determined offline using a branch and bound method;
• receiving data that represents the current parts to be processed by the machine tool workstations in the factory;
• processing the received data using the trained neural network to obtain the candidate sequence of the current parts to be processed;
• and
• verifying whether the determined candidate sequence of current parts to be processed will be delivered on time using Little's Law.