Controling operation of machine tools using artificial intelligence
Methods, systems and apparatus, including computer programs encoded on computer storage medium, for controlling operations of machine tool workstations. Machine tool workstations are grouped into functional groups. Neural networks corresponding to the functional groups are trained to process respective inputs representing parts to be processed to generate respective outputs representing sequences of ordered subsets of the parts that produce a reduced setup time for workstations in the functional groups. Data representing respective collections of parts to be processed by workstations included in the functional groups is processed using the trained neural networks to generate corresponding sequences of ordered subsets of the collection of parts. Average delay times associated with the generated sequences of ordered subsets of the collection of parts are computed. If the average delay times are less than a predetermined threshold, parts are released to the functional groups for processing according to the generated sequences.
1. A method for processing a plurality of parts using a plurality of machine tool workstations, the method comprising:
grouping the plurality of machine tool workstations into multiple functional groups, wherein each functional group includes workstations of the same type;
for each of the multiple functional groups:
training a neural network to process a data input representing multiple parts to be processed by the type of workstations included in the functional group and generate a data output representing a sequence of ordered subsets of the multiple parts that, when processed by the functional group, produces a reduced set up time for the workstations in the functional group;
processing, through the trained neural network, data representing a collection of the plurality of parts, the collection of the plurality of parts comprising parts to be processed by the type of workstations included in the functional group, to generate a corresponding sequence of ordered subsets of the collection of the plurality of parts that, when processed by the functional group, produces the reduced set up time for the workstations in the functional group;
computing an average delay time associated with the generated sequence of ordered subsets of the collection of the plurality of parts;
determining whether the average delay time is less than a predetermined threshold; and
in response to determining that the average delay time is less than the predetermined threshold, releasing one or more parts according to the generated sequence of ordered subsets of the collection of the plurality of parts to the functional group for processing.
2. The method of claim 1 , wherein the plurality of machine tool workstations comprise one or more of (i) cutting tools, (ii) drilling machines, (iii) lathes, (iv) boring machines, or (v) grinding machines.
3. The method of claim 1 , wherein each workstation is associated with a set of performance parameters, the set comprising workstation setup time and part delivery time.
4. The method of claim 1 , wherein workstations of a same type comprise workstations with a common Kanban inventory.
5. The method of claim 1 , wherein each functional group is configured to receive a constrained number of days of work in progress per batch of parts, wherein the number of days depends on an average setup and machining time per part over each workstation in the functional group.
6. The method of claim 5 , wherein a quantity of multiple functional groups is dependent on the computational capabilities of the neural networks and properties of the plurality of machine tool workstations.
7. The method of claim 6 , wherein properties of the plurality of machine tool workstations comprise one or more of (i) location of machine tool workstation, (ii) an acceptable uninterrupted machine tool workstation runtime.
8. The method of claim 1 , wherein, for each of the multiple functional groups, training the neural network comprises training the neural network on a plurality of training data sets, each training data set comprising the data input representing the multiple parts to be processed by the type of workstations included in the functional group and the corresponding sequence of ordered subsets of the multiple parts that, when processed by the functional group, produces the reduced set up time for the workstations in the functional group, wherein the corresponding sequence of ordered subsets of the multiple parts has been determined using a branch and bound method.
9. The method of claim 8 , wherein training the neural network on the plurality of training data sets comprises, for each training data set:
processing the data input representing multiple parts in accordance with current values of parameters of the neural network to generate the predicted sequence of ordered subsets of the multiple parts; and
adjusting the current values of the parameters of the neural network based on the predicted sequence of ordered subsets of the multiple parts and the sequence of ordered subsets of the multiple parts determined using the branch and bound method.
10. The method of claim 1 , wherein the neural network is implemented by cloud computing.
11. The method of claim 1 , wherein a size of the sequence of ordered subsets is dependent on acceptable uninterrupted machine tool workstation runtimes.
12. The method of claim 1 , wherein releasing parts according to the generated sequence of ordered subsets of the collection of the plurality of parts to the functional group for processing generates processed parts, and wherein the method further comprises routing one or more subsets of the processed parts to other functional groups for further processing.
13. The method of claim 1 , wherein determining the average delay time comprises applying Little's Law.
14. The method of claim 1 , wherein determining whether the average delay time is less than the predetermined threshold comprises: computing a standard deviation of delay time; adding a multiple of the computed standard deviation to the computed average delay time to generate an adjusted average delay time; and determining whether the adjusted average delay time is less than the predetermined threshold.
15. The method of claim 1 , further comprising, in response to determining that the average delay time exceeds the predetermined threshold: iteratively reducing a size of the ordered subsets in the generated sequence of ordered subsets until the average delay time is less than the predetermined threshold.
16. The method of claim 15 , wherein a size of the ordered subsets is equal to one, and wherein the method further comprises releasing a random sequence of parts to next available machine tool workstations.
17. A system comprising:
a collection of machine tool workstations used to process a plurality of parts; one or more computers in data communication with the collection of machine tool workstations;
a computer-readable medium coupled to the one or more computers having instructions stored thereon which, when executed by the one or more computers, cause the one or more computers to perform operations comprising:
grouping the plurality of machine tool workstations into multiple functional groups, wherein each functional group includes workstations of the same type;
for each of the multiple functional groups:
training a neural network to process a data input representing multiple parts to be processed by the type of workstations included in the functional group and generate a data output representing a sequence of ordered subsets of the multiple parts that, when processed by the functional group, produces a reduced set up time for the workstations in the functional group;
processing, through the trained neural network, data representing a collection of the plurality of parts, the collection of the plurality of parts comprising parts to be processed by the type of workstations included in the functional group, to generate a corresponding sequence of ordered subsets of the collection of the plurality of parts that, when processed by the functional group, produces the reduced set up time for the workstations in the functional group;
computing an average delay time associated with the generated sequence of ordered subsets of the collection of the plurality of parts;
determining whether the average delay time is less than a predetermined threshold; and
in response to determining that the average delay time is less than the predetermined threshold, releasing one or more parts according to the generated sequence of ordered subsets of the collection of the plurality of parts to the functional group for processing.