Self-learning manufacturing using digital twins

Patent metadata

Field	Value
Publication number	US-2023022733-A1
Application number	US-202117380759-A
Country	US
Kind code	A1
Filing date	Jul 20, 2021
Priority date	Jul 20, 2021
Publication date	Jan 26, 2023
Grant date	—

Abstract

Systems, methods, and computer programming products for self-learning order dressing rules applied to manufacturing products in accordance with received product specifications. The translation from commercial characteristics to manufacturing characteristics of the product being manufactured are learned and adjusted to meet the specifications for quality required by the provided commercial characteristics. Reinforcement learning models learn from the quality characteristics of produced products by applying positive scores when the commercial to manufacturing characteristic translation is on-specification, otherwise a penalty is applied when an off-spec product is produced. Digital twins of manufacturing equipment, simulated in real time, provide insight and recommendations for achieving correct quality characteristics. Sensors in each device or within the surrounding environment help digital twins to measure operational performance and lifecycle of the manufacturing equipment against historical baselines. Reinforcement models dynamically adjust equipment settings for producing products to account for equipment performance degradation over time and changes in operation performance.

First claim

What is claimed is:

1 . A computer-implemented method for dynamically augmenting manufacturing characteristics configured to produce an on-specification product, the method comprising:
developing, by a processor, a reinforcement learning model using a historical rules database comprising existing rules translating commercial characteristics of a product into the manufacturing characteristics, and historical production settings comprising quality parameters achieved during historical production of on-specification products in accordance with the existing rules; inputting, by the processor, a set of commercial characteristics to produce the product into the reinforcement learning model; receiving, by the processor, output from the reinforcement learning model translating the set of commercial characteristics into manufacturing characteristics for producing the product; recording, by the processor, quality characteristics of product produced by one or more manufacturing systems and production settings of the one or more manufacturing systems used to produce the product; comparing, by the processor, the quality characteristics of the product produced by the one or more manufacturing systems with the commercial characteristics of the product; and rewarding, by the processor, the reinforcement learning model for correctly translating the commercial characteristics to the manufacturing characteristics, whereupon the quality characteristics of the product match the commercial characteristics.

2 . The computer-implemented method of claim 1 , further comprising:
penalizing, by the processor, the reinforcement learning model for incorrectly translating the commercial characteristics to the manufacturing characteristics, whereupon the quality characteristics of the product do not match the commercial characteristics.

3 . The computer-implemented method of claim 2 , further comprising:
analyzing, by the processor, changes in operating conditions of the one or more manufacturing systems producing the product; and modifying, by the processor, the reinforcement learning model to predictively compensate for the changes in the operating conditions of the one or more manufacturing systems.

4 . The computer-implemented method of claim 3 , wherein the changes in the operating conditions of the one or more manufacturing systems producing the products are measured by sensor data collected by one or more sensors onboard the one or more manufacturing systems or positioned within a surrounding environment of the one or more manufacturing systems.

5 . The computer-implemented method of claim 4 , further comprising:
tracking, by the processor, the sensor data in real time; inputting, by the processor, the sensor data into a digital twin, and reflecting the changes in the operating conditions of the one or more manufacturing systems in the digital twin; assessing, by the processor, performance and lifecycle of the one or more manufacturing systems as a function of the changes in the operating conditions as measured by the sensor data using the digital twin to simulate the one or more manufacturing systems under the changes to the operating conditions and comparing simulation results to historical baseline performance of the one or more manufacturing systems using the historical production settings; and recommending, by the processor, an adjustment to the settings of the one or more manufacturing systems or the existing rules used to translate commercial characteristics of a product into the manufacturing characteristics to compensate for changes in the operating conditions in order to achieve correctly translated manufacturing characteristics from the commercial characteristics.

6 . The computer-implemented method of claim 1 , wherein the manufacturing characteristics include mechanical properties, chemical properties, tolerance properties, grade, or dimensions of the product.

7 . The computer-implemented method of claim 3 , further comprising:
capturing, by the processor, one or more metrics indicating degradation of equipment comprising the one or more manufacturing systems; inputting, by the processor, the one or more metrics into the reinforcement learning model; and adjusting, by the processor, the existing rules of the reinforcement learning model to compensate for the degradation of the equipment manufacturing the product.

8 . A computer program product comprising:
one or more computer readable storage media having computer-readable program instructions stored on the one or more computer readable storage media, said program instructions executes a computer-implemented method comprising:
developing, by a processor, a reinforcement learning model using a historical rules database comprising existing rules translating commercial characteristics of a product into manufacturing characteristics, and historical production settings comprising quality parameters achieved during historical production of on-specification products in accordance with the existing rules;
inputting, by the processor, a set of commercial characteristics to produce the product into the reinforcement learning model;
receiving, by the processor, output from the reinforcement learning model translating the set of commercial characteristics into the manufacturing characteristics for producing the product;
recording, by the processor, quality characteristics of product produced by one or more manufacturing systems and production settings of the one or more manufacturing systems used to produce the product;
comparing, by the processor, the quality characteristics of the product produced by the one or more manufacturing systems with the commercial characteristics of the product; and
rewarding, by the processor, the reinforcement learning model for correctly translating the commercial characteristics to the manufacturing characteristics, whereupon the quality characteristics of the product match the commercial characteristics.

9 . The computer program product of claim 8 , further comprising:
penalizing, by the processor, the reinforcement learning model for incorrectly translating the commercial characteristics to the manufacturing characteristics, whereupon the quality characteristics of the product do not match the commercial characteristics.

10 . The computer program product of claim 9 , further comprising:
analyzing, by the processor, changes in operating conditions of the one or more manufacturing systems producing the product; and modifying, by the processor, the reinforcement learning model to predictively compensate for the changes in the operating conditions of the one or more manufacturing systems.

11 . The computer program product of claim 10 wherein the changes in the operating conditions of the one or more manufacturing systems producing the products are measured by sensor data collected by one or more sensors onboard the one or more manufacturing systems or positioned within a surrounding environment of the one or more manufacturing systems.

12 . The computer program product of claim 11 further comprising:
tracking, by the processor, the sensor data in real time;
inputting, by the processor, the sensor data into a digital twin, and reflecting the changes in the operating conditions of the one or more manufacturing systems in the digital twin;
assessing, by the processor, performance and lifecycle of the one or more manufacturing systems as a function of the changes in the operating conditions as measured by the sensor data using the digital twin to simulate the one or more manufacturing systems under the changes to the operating conditions and comparing simulation res

Assignees

• IBM

Inventors

• Kaul Abhishek
• Ray Rajesh
• Nagar Seema
• Vadayadiyil Raveendran Pramod

Classifications

• G05B19/41865 Primary

  characterised by job scheduling, process planning, material flow · CPC title

• G05B2219/31352

  Expert system integrates knowledges to control workshop · CPC title

• G05B2219/32254

  Work sequence, alternative sequence · CPC title

• G05B2219/35009

  Dynamic simulation · CPC title

• G05B13/028

  using expert systems only · CPC title