Why AI in manufacturing? Competitiveness ultimately drives any company, especially in manufacturing. Hence, it is no surprise that manufacturing companies are now investing more resources in AI for automation and complex analytics. We believe that AI is a force multiplier on technological progress, and specifically in the manufacturing competitiveness. According to McKinsey Global Institute, manufacturing stands to benefit the most from AI, and specifically in applications such as predictive maintenance. AI’s ability to make sense from data, including audio and video, means it can quickly identify anomalies to prevent breakdowns, whether that be an odd sound in an engine or a malfunction on an assembly line detected by sensors.
Looking at and interpreting data generated during the manufacturing process to find ways to reduce waste, improve quality and increase yield is not new. However, the increased use of digital technologies in manufacturing is changing the analytical landscape. Data is generated by digital sensors and actuators, connectivity of machines and the manufacturing environment, and many other sources generally described as the Internet of Things (IoT). This explosion of data increases the complexity for humans to find patterns and trends in the masses of data.
Traditionally, manufacturers have financed improvements as capital expenditures. AI offers a lower cost alternative by enabling companies to use software to analyse the vast amount of data available in and around the factory. AI enables the manufacturer to get more out of its heavy assets by improving throughput, reduced energy consumption, and better process and quality control.
What is AI?
AI is the term that encapsulates machine learning and deep learning. Machine learning is used to ‘learn’ how to execute a specific task from data. Image-recognition is one example where a machine is given thousands of pictures to analyse and acquire the ability to recognise patterns, shapes, faces, and more (based on the features extracted). In machine learning, you need to choose for yourself what features to include in the model, e.g. which features represent the data best to execute the task. Deep learning is an advanced form of machine learning that mimics the layers of neurons in the brain (neural network) to build up a complex AI model, similarly how a small child learns from experience – (s)he would recognise a chair even it does not have four legs, a seat and a backrest. You do not need to understand which features provide the best representation of the data; the deep neural network learns how to select critical features, which are then used to ‘learn’ how to execute a task.
AI and robotics are often used interchangeably, but are two different concepts. AI takes robotics (virtual or physical) and automation to the next level by introducing human-like cognitive abilities. Robots with AI can act beyond rules. AI enables learning in processes and machines thereby making them more effective for the task at hand, using data to modify and adapt behaviour, especially where tasks are repetitive.
Using traditional statistical methods (with the aim to analyse and summarise data) is very dependent on the user and his/her capability to identify possible patterns and trends in data; constrained assumptions are made about the problem and data distributions. AI, using machine learning and deep learning techniques, can analyse massive amounts of data to better understand what the outcomes might be for all (or at least many more) possible options; no firm pre-assumptions about the problem and data distributions are made as the goal is to learn from the data.
Use cases
In operations, machine learning is applied to enable supply chain and inventory optimisation. This includes prediction of optimal stock levels, best routes for delivery and collections, warehousing and other logistics; all leading to cost reduction and improved productivity. Predictive quality management is applied to identify quality issues earlier in the production process leading to a reduction in waste, improvements in quality and an increase in yield.
Predictive analytics provide new ways to better understand when equipment may fail; when it requires maintenance, repair or replacement (in part or whole). This impacts directly on the supply chain by decreasing the levels of stock to be carried ‘just in case’ something breaks. Another way to use predictive analytics is to predict demand for products, even products that are not currently manufactured (new pipeline). Augmenting the manufacturing process data with customer behavioural data can provide a further competitive edge, for example, customisation of goods to specific customer needs.
Advanced business and economic analytics, sales demand forecasting, and predicting the ideal time to purchase material and parts considering exchange rate fluctuations are examples where AI is applied. Intelligent process automation is used to alleviate the burden on administrative personnel to perform tedious or manual tasks which in turn improves administrative productivity in the organisation.
AI in marketing and sales significantly improves the customer experience and effectiveness of marketing efforts. Machine learning techniques applied to cluster an organisation’s customer base enables personalised and directed marketing, automated personalisation of products and services, and prediction of customer needs. The company’s brand is protected through automated monitoring and sentiment analysis of emails and social media; personalised customer engagement is thus possible over multiple channels including conversational online assistants. The effectiveness of sales efforts is assessed, and sales trends and patterns are predicted near real-time through machine learning techniques.
Certain health and safety concerns are effectively addressed through the application of AI. High-risk situations can be identified, and incidents predicted enabling managers to mitigate risks and prevent incidents. Compliance to health and safety regulations are monitored through real-time video and image analysis.
Other operational efficiencies can be obtained from AI in energy management and resource optimisation.
Practical AI implementation considerations
Data
Appropriate data is important. If sufficient data is not available, data can be sourced by adding IoT, integrating with other business and surveillance systems, or accessing external information. Master data management can be applied to assist in understanding and managing data at a macro level.
Clean data is important. Many successful AI projects entailed significant effort in cleaning the data before machine learning can be applied.
Computing power
Significant processing capability (computing power) is used in building and training the initial AI models. Deployment of the AI system will require computing power to continuously refine the models in the operational environment. This should be addressed once as part of project design and implementation.
Expertise and methodologies
Access to AI expertise: staff with the relevant experience in applying machine learning techniques to manufacturing problems are in short supply. Work with a dedicated AI team, often located at an industry partner and use methodologies specific to developing AI implementations, including AI readiness assessments, AI roadmaps, model development, etc. Focus on understanding the business, stakeholder/customer needs, and opportunities to maximise the value from the AI implementation.
Threats
• Will AI create unemployment? This is not a new fear. In the beginning AI will eliminate some of the human tasks; we need to find ways to adopt and re-skill ourselves. Then it has potential to create more jobs than it eliminates. This is maybe somewhat similar to transition from horses to cars during the first industrial revolution. Similarly, when ATMs or computers came around in the 70s and 80s. Like with other revolutions, AI adoption requires mastery of new skills.
• Bias: AI models are created with the data at hand. If the data is not fully representative of the problem space, the model will be biased and will not provide fair and reasonable decisions for all instances where it is used. AI can be used to influence perceptions which ultimately can negatively impact on possible decisions. Therefore, AI implementation needs to be cognisant of potential bias and ensure representative data.
• Governance and ethics: software glitches could easily cause AI mistakes. There needs to be clear accountability and regularity ownership, e.g. who is responsible if a self-driving car or a drone makes a severe accident? Society needs to ensure that our complex AI systems do what we want them to do.
Where are you on the AI journey?
An AI readiness assessment will map out where you are on the AI journey, and where AI can have the biggest impact. The following describe broad categories of where organisations may find themselves in their AI journey:
• You have not yet implemented any digital technologies. You stand to gain the most from this journey towards AI. Implementation of a cost-effective digitalisation program aligned to your business strategy will lay a solid foundation for your AI journey. This can be achieved through the introduction of IoT as an example.
• Your factory is equipped with many sensors and systems to generate lots of data (already IoT enabled). You are ready to fast track your AI journey. Your opportunity is in the integration of your data in process management and control. The challenge is to identify and address the missing data that will get you the biggest leverage from AI. Early AI implementations can already start.
• You have implemented a digital strategy. Your data is integrated with integrity and complete. You are ready for full-scale AI implementation with continuous improvement.
Conclusion
In summary, adopting AI will lower cost, improve yield and empower the manufacturer to provide distinct value to customers. As part of mainstreaming AI, various use cases provide evidence of the improvements generated using machine learning and deep learning.
Many of the exceptional companies are embracing AI fully – for any organisation that have not yet taken AI seriously, now is the best time to start getting up to speed.
Partners with capabilities in AI can collaborate with manufacturers to implement AI. We propose that an AI readiness assessment be performed, and an AI roadmap developed to guide implementation and investment to accelerate technological progress and manufacturing competitiveness. Early implementation of AI use cases can start in parallel to AI readiness assessments.
© Technews Publishing (Pty) Ltd | All Rights Reserved
printer friendly version