In recent months, businesses in all sectors have started exploring how to incorporate AI into their processes; the manufacturing industry is no exception.
As Mohammed Salifu (pictured), Group Data and Analytics Director, Domino Printing Sciences, outlines, AI offers enormous potential ... but its successful adoption demands careful planning, particularly around data collection and curation.
What is artificial intelligence?
AI describes any application of computer software that allows machines to mimic human intelligence to enable problem-solving — be this with vision, speech or interpretation of data.
It’s an umbrella term that describes several methodologies, including robotics, image analysis, language processing, machine learning and artificial neural networks.
On a basic level, artificially intelligent systems identify patterns through algorithmic data analysis. More complex systems can learn from experiences, solve problems and make decisions without human intervention.
Today, AI applications are in use across a wide range of industries:
- Food and beverages: Campbell Soup Company uses AI to analyse consumer preference data and agile design methodology to accelerate the development of new products
- Waste and recovery: Greyparrot, a company specialising in AI-generated waste analytics has developed computer vision systems for waste identification at materials recovery facilities
- Coding and marking: Domino, incorporates aspects of AI to target values for new formulations and automate testing to speed up the ink development process.
AI in the manufacturing industry
There are three key areas where AI is proving valuable in the manufacturing industry:
- Error reduction: AI systems can be developed to understand and analyse all types of visual data, including data from quality control systems on production lines to identify patterns that could indicate wider production issues, and facilitate waste and error reduction
- Predictive maintenance: data from maintenance logs and production line performance can be used to predict machinery performance and when parts need replacing or maintenance is required.
- Forecasting: with a thorough dataset including information on plant operations, production performance, and sales and feedback, AI systems can forecast demand, helping manufacturers streamline inventory and preplan production runs.
Getting started with AI: data first
Any use case for AI – whether in manufacturing or any other sector – requires a dataset large enough to train an AI model. Indeed, data is the first step in any AI journey and is arguably the essential part of the process, without which attempts to implement AI are destined to fail.
As such, manufacturers must implement systems and processes that enable consistent, reliable data collection across all necessary production activities before getting started. The existence of a joined-up governed data architecture is a prerequisite to a successful AI implementation.
Quality control: AI applications lend themselves particularly well when used alongside machine vision systems for quality control. Visual quality control systems can facilitate data collection to build the datasets required to train AI models.
The same systems could also become visual input sources for analysis and decision-making, feeding directly into AI models to process the data and extract insights.
Machine metrics: Robust data production equipment will be vital in enabling AI for predictive maintenance. Manufacturers can collect valuable insights on machine performance and diagnostics via cloud-based monitoring solutions.
Historical data can be used to train AI models, whereas AI algorithms can analyse real-time machine data to predict when maintenance is needed.
Production data: Wider production data, encompassing all parts of the production line, will be required for AI in performance optimisation, predictive maintenance, and forecasting. By integrating equipment, manufacturers can collect production data from the plant floor and consolidate it into an accessible dataset to support the deployment of AI.
Upstream and downstream data: When combined with other production monitoring systems, variable data coding at the batch or item level can be used to tie individual products back to the production line.
A serialised product code will allow the identification of products if they land in a reject pile or cause an issue at some point during distribution, providing a route to trace back and uncover precisely when and where they were made.
The value of variable data codes can extend far beyond the factory as products move through the wider supply chain and into the hands of consumers. A scannable code with a unique serial number can be used to gather customer feedback and associate it back to the product’s unique production history.
This traceability not only helps with identifying where issues arise but can also help brands to collect data on consumer preferences and trends – which can help to inform the development of existing or new products.
Collecting this information during production and beyond the factory doors is another part of a complex toolkit to help manufacturers get to a point where their data is robust enough to consider investigating AI applications.
Preparing a workforce for AI
Preparing for an AI manufacturing project will require sufficient resource allocation to implement new systems, develop datasets, train AI models, and monitor and analyse progress.
Although conversations surrounding AI inevitably bring up concerns about the replacement of human workers, for the short-term at least, the opposite is true.
Forbes suggests that AI will enable workers to focus on more meaningful and high-value activities. MIT and Statista suggest that human-robot collaborations – or cobots (which can be up to 85% more productive than teams made of either humans or robots alone) — will be the future of manufacturing.
Preparing a workforce for AI will be an ongoing process. As technologies evolve, businesses must invest in learning and development to ensure employees remain equipped with the skills necessary to progress.
Conclusion
The impact of AI on manufacturing is likely to be substantial, but it’s not a sure-fire route to success.
A strategic plan should start with understanding the use case for AI, facilitating reliable data collection methods aligned to that use case, discussing requirements with existing solutions providers to discover what data is already available and what solutions can facilitate seamless data collection, and ensuring your workforce is part of your AI journey.