Machine learning to shift production lifecycles

Automating factories has been a continuous process over the last few decades as machinery is updated regularly, but with the advent of Industry 4.0 the evolution of these technologies has triggered the convergence of operational technology with information technology.

AI employs the use of machine learning, which promises a new era in manufacturing, a disruption that will reap many benefits, enhancing efficiencies and productivity. While machine learning is perhaps at the early phase of the adoption curve, it is showing signs that it will dramatically transform manufacturing and shape the factories of the future. Gathering data has become a normal function in any automated factory but making sense of it without the support of analytics and AI is not possible. Data from automation systems such as PLCs, DCS and SCADA is now getting the attention it deserves, as the data that has been gathered can now be categorised and used to enhance efficiencies and optimise output using AI.

It is early days for manufacturers in the UK for adoption of AI, but the current COVID-19 situation will add to this momentum of quicker adoption. Whilst it is premature, there are some industries like aerospace that have embraced machine learning with great results. We are also seeing an upsurge of this in FMCG production and there will be a much higher uptake in the chemicals, F&B and logistics industries in the near future.

Currently, smart simple methods are used in machine learning. For instance, in pharmaceutical and F&B manufacturing, use of cameras to detect poor quality of the product in all its stages has been a successful practice.

A good example is that of bread production in a factory where monitoring the quality has traditionally been a manual process. Today parameters can be monitored in real time and streamlined using trained machine learning algorithms.

Proof of concepts

Importantly, using proof of concepts with machine learning by comparative models can be identified to monitor the changes. These models are then validated against the key performance indicators until the preferred model becomes a standard.

Key performance indicators are quality, production time, temperature control, variability, etc. and these vary as per the product being manufactured. Additionally, management can also use machine learning to ensure that production costs are maintained, leading to maximum returns on investment.

Manufacturers will see strategic shifts as machine learning becomes ubiquitous in factories. These changes will be seen in the way users interact, allocate, manage and view their entire production process. Both IT and OT domains in manufacturing will converge and investment in AI and related technologies will increase. To remain robust, the aim will be to have measurable goals with shorter implementation cycles to deliver greater end results in a much shorter time.

AI algorithms are not new but scaling them up and lowering the barrier to entry to incorporate them into the automation of factories to implement practical solutions in established manufacturing plants is. So, the way these technologies are adopted can also vary, and whilst internet of things (IoT) and cloud-based solutions have created ripples in industry they may not be the ultimate solution for all manufacturers. For most shop floors, local intelligent edge devices working together in network allows smart devices to work collaboratively together and with the cloud where required to deliver the speed of response often critical to manufacturing applications. This holds far greater value and is seen as the direction that most manufacturers are heading towards.

As providers of cutting-edge technologies in manufacturing, it is natural for us to work with business partners in co-creating solutions that work for them and yield best results that are sustainable. For instance, using cloud technology may not always be ideal for key industries, such as chemicals and pharmaceuticals, where high latency to retrieve data and respond is not acceptable. So, working with a self-contained platform which has low latency can allow each platform, cloud or edge to deliver value that plays to their strengths.

Of course, there are challenges too, such as taking these specialist tools and putting them in the hands of technicians and engineers who may not be trained to manage the complex new technologies. So, we work with industry organisations, government catapults and academia to build these skills as well as working on standard tools to support the deployment and operation of AI on the factory floor.

Robot control

When we talk about AI, we cannot ignore how robotics comes to prominence in a smart factory of the future. And it is most useful when there are common tasks that an algorithm can quickly learn and optimise such as visual quality inspection and grasping detection. In such instances using machine learning in combination with automation technology to control robots is highly beneficial.

The ability of robots to carry out repeatable functions also means freeing employees to do much more meaningful tasks that are further effective. As technology providers, manufacturers are looking towards vendors who can provide these common platforms and comprehensive portfolios that are open enough to integrate with existing plant equipment to produce a truly agile production facility.

The benefits of machine learning are far-reaching and manufacturing as a sector is only currently scratching the surface. Not only does use of AI help drive production processes at all levels of manufacturing, it can be applied to deliver end-user benefits too. Manufacturers can generate new business models, and many have successfully used data analysis to create successful brands in other areas of business, such as finance and the service industry. Two very prominent examples are Ocado and Amazon who started by optimising logistics and quickly moved to become core technology providers in their own right.

It is not realistic for every manufacturer will be able to employ data scientists and machine learning specialists, but with industry offerings available, along with open standard hardware available from suppliers, manufacturers will be able to outsource these tasks to buy in self optimising solutions.

Machine learning will quickly move from a ‘nice to have’ solution for the future to an absolute necessity in a short period of time. It will also be the future of how collaborative relationships are developed between workers and machines on the factory shop floor. And while machines will take over some types of jobs new skills will be required to work and maintain these systems creating new roles and disciplines. As much as the machines will be developed to provide artificial intelligence in a factory, in the production life cycle they will not replace human emotions like empathy, thought and the power to make a decision.