While popular Industry 4.0 technologies tend to focus on robotics and the Internet of Things, less tangible technologies such as artificial intelligence are poised to make as much of an impact. By Jonathan Chou
The trend for increased automation and data exchange can certainly be seen today with manufacturers aiming for a modularised “smart factory” incorporating the Internet of Things (IoT), cloud computing and cyber-physical systems.
This signifies “a new shift in the production and manufacturing processes, drawing away from static production chains, to a more flexible, individualised and efficient idea of production,” according to M Dopico in his paper “A Vision Of Industry 4.0 From An Artificial Intelligence Point Of View”.
Opportunities For Emerging Countries
By importing Industry 4.0 technologies and combining them with low costs of production factors, emerging countries such as Malaysia or China can leapfrog in the value chain, according to market research provider Euromonitor.
However, Mr Dopico believes that the new Industry 4.0 environment will benefit from a system that is able to analyse, filter and interpret huge amounts of information from different types of sensors, in order to suggest the most recommended course of action. Most of these concepts share the same background theory as artificial intelligence (AI)—and the potential for the technology to affect manufacturing has not gone unnoticed.
Not A Small Deal
While popular perceptions of the “smart factory” tend to zoom in on industrial robotics or the IoT, AI is poised to make as large of an impact on manufacturing and the metalworking world.
As much as US$15.7 trillion could be contributed by AI to the world economy by 2030, with US$6.6 trillion coming in the form of increased productivity as businesses automate processes and augment their labour forces with new AI technology.
The Pricewaterhousecoopers report that shared these findings also indicated that global gross domestic product, which stood at about US$74 trillion in 2015, will be 14 percent higher in 2030 as a result of AI.
Future Deployment By Machine Tool Builders
The European Association of Machine Tool Industries (CECIMO) certainly believes in the potential of the technology. The association represents the common interests of the machine tool industries, which includes approximately 1,300 industrial enterprises in Europe, and covers 98 percent of total machine tool production in Europe and around 36 percent worldwide.
The association held its general assembly in November last year to address the challenges of AI, and to develop the right framework conditions through cooperation with decision makers in the European Union and industry stakeholders.
The move is aimed at the deployment of AI and machine learning by machine tool builders in the coming years. “Given that the enhancement of production efficiency on a continuous basis has been an everlasting commitment for both our industry and our customers, both artificial intelligence and machine learning offer a tremendous opportunity for automation, increased engineering efficiency and reduced costs,” said Luigi Galdabini, CECIMO president.
Machine tool consumption forecasts for 2018 by CECIMO suggest a 4.1 percent rise in Europe and an increase of 3.6 percent in Asia and the rest of the world.
The investments in modern machine tools are fuelled by investment tax incentives in certain countries and the trend towards digital manufacturing.
AI In CNC Machining
The optimal cutting parameters can lower machining errors such as tool breakage, tool deflection and tool wear, resulting in higher productivity at lesser cost. But the question remains: Can AI be used to optimise the metal cutting process in real-time?
Research from Kyung Sam Park and Soung Hie Kim from the Korea Advanced Institute Of Science And Technology aims to tackle that question.
The researchers identified three main categories of AI-based methods for controlling computer numerical control (CNC) machining parameters:
- Knowledge-based expert systems (KBES) approach.
- Neural networks approach.
- Probabilistic inference approach.
These approaches represent different online methods to automatically adapt and optimise machining parameters based on sensor information on machining responses in real-time, without any computer simulations beforehand.
Three components are required for online control of machining parameters, of which an AI-based approach is used to handle the second and third components:
- Sensing devices.
- Information representation from the sensors.
- Optimising machining parameters.
Mr Kyung and Mr Soung added that a simplistic adaptation of AI techniques to machining control would be inadequate. This is because execution time of these systems is generally too long when compared to the reaction time required for the machining control, particularly if it is on a complicated workpiece.
A KBES approach intends to use advanced computer programs to mimic the problem-solving strategies of human experts to provide recommendations. The KBES can capture causal and inferential knowledge about machining processes to provide expert-level recommendations. In theory, this would be a valuable aid to machining operators, who are facing increasingly complex tasks.
The structure of the KBES consists of three modules:
- Knowledge base.
- Inference engine.
- Sensor data acquisition and processing module.
The inference engine drives the system and interfaces with the knowledge base to provide advice to the operator. With a database of near-optimal machining control, the knowledge base can provide explanations to justify the KBES’ line of reasoning.
In order to achieve the near-optimal machining control such as tool changes and machining stops, an adaptive control algorithm is used to determine various constraint rules.
Unlike KBES, neural networks generate their own knowledge by learning from examples. This means that neural networks can rapidly generate their own knowledge implementing a learning rule. Costly and time-consuming knowledge acquisition processes used in the KBES can then be skipped entirely.
Learning is supervised, in which connection weights of the network are adapted in response to the inputs and desired output pairs. Neural computing is also both distributed and associative, which means that multi-layer neural networks are able to register important features in their hidden layers.
A trained neural network for machining could use this “hidden knowledge” to generate important generalisations, allowing for a reasonable response even when presented with incomplete or previously unseen data. An example of a neural network for machining knowledge is shown below.
Probabilistic Inference Approach
Another AI-based approach uses influence diagrams specially developed for making complex decisions based on information from a variety of sources. Key variables are then identified with influences or relationships between each other, and represented as nodes:
- Chance nodes represent highly variable values such as cutting force, tool wear, and acoustic emissions.
- Decision nodes are values that are chosen by the operator, such as feed rate or cutting speed.
- Value nodes represent the objective to be maximised. This would most likely be the metal removal rate.
Once a complete influence diagram has been generated, different aspects are manipulated and evaluated until an optimal decision strategy has been reached.
An example of an influence diagram for machining optimisation is shown below.
Rising AI Interest
Machining technologies infused with AI have already begun making its way into the market. Machine tool maker Okuma launched the CNC control OSP-P300A in Japan last year, which comes with AI integrated within the control’s central processing unit.
The control can diagnose axis feed failure and the presence or absence of ball-screw or ball-screw support bearing malfunctions. The company also stated that it could also identify and locate abnormalities that are often difficult to resolve.
Another machine tool maker, DMG Mori, stated in 2017 that one of the most important AI applications in the machine tool industry should be predictive maintenance, as it “requires accumulated data, as well as factor analysis of each error and development of according preventive measures”.
The company added that they have gathered comprehensive data through more than 300,000 installed machines in the global market, and have been exploring causes of machine troubles and the solutions through their product problem report system for more than 20 years.
Along with their European group company AG’s presence and experience in the 5-axis machine market, they believe that they “are capable of developing an accurate preventive maintenance mechanism with AI as a pioneer in the machine tool industry”.
A new advanced technology centre was also opened at its global headquarters in Tokyo. Employees aged 35 or younger will be selected to start working as researchers at the technology centre, where they will spend two years obtaining a range of specialist qualifications related to artificial intelligence and industrial IoT technologies. The move is aimed to develop staff that will lead open-innovation initiatives or technologies that cross over to the information technology industry.
“The world is well on its way to an AI-powered revolution—where cognitive systems will truly augment human capabilities, but at machine speed and big data scale,” said Amir Husain, founder and chief executive officer at SparkCognition.
The technology company is developing a self-learning, data-driven analytics platform for the safety, security and reliability of data technology for manufacturers in the aerospace, oil and gas, defense, and energy industries.
With risk mitigation and efficiency being the top priority for in the increasingly competitive global market, making decisions at “machine speed and big data scale” seems to be an attractive proposition indeed.