Mapping the Application of AI to Manufacturing

AI is often touted as the “next big thing” in manufacturing, after the Industry 4.0 revolution. Where will it come from, where will it be implemented, what impacts will it have?

After decades in which AI was always 5-10 years from being implemented in the real world, in the past decade that promise has finally begun to be seen. From computer vision for automatic monitoring of the factory floor [4], to predicting likely outcomes of genetic manipulations in synthetic biology [3] , the potential impact of AI in manufacturing is immense. This evolution of manufacturing will involve a variety of underlying AI technologies (computer vision, various forms of machine learning, intelligence support for human decision-making, etc.) as well as targetted applied research for specific types of application. In addition to the technical research there is research in business management and economics. The legal, ethical and social implications also need attention. Some industries are more likely to have early adopters than others. Governments will be involved, from funding much of the fundamental research, and some of the applied research and innovation activity, through industrial policy and regulation. An initial overview of this landscape is presented below.

Underlying Technologies

Both software and hardware technologies are involved in the application of AI to manufacturing activity. Some of the most common types of hardware and software are:

• Software
  • Machine Learning
  • Natural Language Processing
  • Computer Vision
• Hardware
  • Robots
  • IoT
  • Cameras

Applications

While almost all aspects of manufacturing could potentially be enhanced by the use of AI, some aspects are more advanced than others in the development of relevant applications than others. Some of those with currently applicable applications available are:

• Machinery Observation
  • Maintenance
  • Safety
• Materials Observation
  • Inputs (raw materials/components)
    • Quality Checks
    • Logistics
  • Outputs (products)
    • Quality Checks
    • Logistics
• Product Design
  • Using Customer Data
  • Using Market Data
  • For Product Utility
  • For Physical Form
    • For Aesthetics
    • For Functional Utility
• Production Planning
  • Using Customer Data
  • Using Production Data
  • Using Supplier Data
• Physical Security
• Cybersecurity

Industries

While all manufacturing industries could potentially benefit from AI applications, some are more likely than others to be amongst the early adopters. These include ones with an already advanced level of automation, particularly early adopters of Industry 4.0 approaches, and those for which the benefits are larger and more easily attained, such as:

• Automobiles, e.g. Industry 4.0 machine data used for real-time monitoring and adjustment of the vehicle assembly production line for efficiency [6];
• Energy/Power, e.g. Machine Learning applied to cable termination bonding design  [1];
• Pharmaceuticals, e.g. Machine Learning for metaboliute identification in drug design [5];
• Heavy Industry (such as Construction equipment manufacturing), e.g. machine vision for early detection of heavy robot manufacturing failures;
• Food and Beverage, e.g. machine vision for product quality and food safety checking.

Mini-Case Study:  Deep Learning for Predictive Maintenance for Food Packaging Equipment

Predictive maintenance involves taking machinery offline and checking parts for wear, replacing lubricants and similar activity before faults occur. As highlighted by the Toyota Production System/Lean Manufacturing model, allowing a machine to fail can be very costly by shutting down a whole production line unexpectedly. Traditional predictive maintenance tends to use a conservative (short) cycle for maintenance, which involves higher costs up-front in order to avoid the much larger costs of a production line shutdown. Certain industries have particularly high costs if the production line shuts. One of these is the fresh food packaging industry where whole batches of produce may spoil somewhere in the supply chain if the packaging machinery unexpectedly shuts down. Lengthening the time between predictive maintenance periods while maintaining very low risks of machine failure can drastically improve efficiency. In 2019 [2] reported a robust case study in this area showing the potential benefits of such a system on time-sensitive production lines.

Mini-Case Study:  AI in Drug Design Safety

When drugs are introduced into a human body, as well as the intended impact on the target ordinary processes in the body alter the chemical(s) in the drug into different chemicals (referred to as metabolites). These new chemicals may be harmful. During the drug design process, predictions for the potential metabolites are a key component of identifying potentially safe and efficacious drugs. This prediction process is very complicated and still far from completely understood. Some current research on improving metabolite prediction includes the development of machine-learning approaches to broaden the search for metabolites for candidate drugs [5].