The self-learning factory of the future. By Karen Krivaa
Recent innovations in cloud computing and big data storage and analysis are driving a huge increase in the demand for machine learning solutions. Manufacturers are adopting machine learning (ML) models because it makes factories smarter and more efficient, which means higher revenues. According to McKinsey, AI and machine learning have the potential to create an additional $2Tn in value for manufacturing and supply chain planning as processes and decisions for improved productivity become optimised by leveraging big data.
Here are five ways that manufacturers can use machine learning to their advantage:
1. More accurate demand forecasts – All manufacturers are faced with the challenge of forecasting demand accurately to avoid over stocking and shortages. Using machine learning, systems that leverage the correct attributes and can continuously be trained, can help companies plan and continuously adjust production to produce the exact quantities that meet sales requirements while also being able to tweak predictions based on the impact of changing events such as new product introductions and supply chain disruptions. Using data from external sources such as social, news and weather networks can also enhance the prediction accuracy. The end result is less cash tied up unnecessarily in products and materials and fewer delivery delays due to lack of product availability. According to McKinsey, by using machine learning for demand estimates, manufacturers can experience an overall inventory reduction of 20 per cent to 50 per cent.
2. Predictive maintenance – Predictive maintenance fixes small problems before they become bigger ones, reducing lost production time and maintenance costs while increasing equipment life. Equipment is monitored so that it is serviced when needed instead of at scheduled service times, for example, a machine learning model can use a baseline of collected performance data to detect when there is an increase in vibrations that may indicate a malfunction. If the ML model can run on the streaming data simultaneously with historical data, the outcome will be more accurate. Since industrial equipment is very expensive to purchase and maintain, increasing uptime and product reliability while delaying replacements costs results in significant savings. McKinsey estimates that predictive maintenance will generate a ten per cent reduction in annual maintenance costs, up to a 20 per cent downtime reduction and 25 per cent reduction in inspection costs for industrial equipment. In addition, using predictive maintenance can ensure employee safety by reducing work related accidents.
3. Hyper-personalised manufacturing – Machine learning is enabling companies to take personalisation to the next level, evolving from mass production to mass customisation. Personalisation of products is becoming an important differentiator. Both BMW and Mercedes Benz provide car buyers the opportunity to customise the car of their choice online and see what the final product will look before their order is sent to production. Machine learning can be used to identify if a preference for opulence or minimalism can be specific to a generation, region, or culture, and person – based on real-time and historical activities. The 360-degree customer view and the ability to react in the moment can create a positive and unique experience for each customer. Machine learning also enables machinery to identify when products need to be configured differently and then automatically adapt production on the shop floor so that different types of products can be produced on the same assembly line.
4. Optimised production runs – Machine learning algorithms are capable of autonomously improving the efficiency of manufacturing processes by monitoring quantities used, cycle times, temperatures, lead times, errors, and down time. Starting initially as an ‘operator assist’ mode, where systems suggest answers to the operator, they will learn from operators’ decisions and actions to become more and more autonomous, perhaps even eventually replacing operators. In the future, machine learning will work in a vendor agnostic environment where all machines will speak the same language, increasing production efficiency from machine to machine across the entire shop floor.
5. Automated procurement – Analytics combined with machine learning can record and critique every stage of the procurement process. The first step in uncovering sourcing opportunities can transform from a timeconsuming, manual task to a real-time, automatic response. Procurement can apply machine learning to determine the best possible starting rate or negotiations, and also discover the best contract terms that will help the partnership become more successful. Machine learning can pick out hidden patterns that indicate when a supplier is not meeting business and regulatory requirements, departments that are most likely to overspend and suppliers that may face difficulties. Honeywell has already incorporated machine-learning algorithms to improve procurement, strategic sourcing and cost management.
In order to become smarter and learn from previous outcomes, machine learning models need to rapidly ingest, process and analyse, huge volumes of both streaming and historical data with ultra low latency. Models that can’t process streaming and historical data fast enough can result in less accurate and timely results. Distributed In-Memory computing architectures can speed up analytical and transactional processing and provide scalability, especially at peaks by running services and analytics on a unified big data speed layer, resulting in lower processing overhead and higher data quality. In-Memory computing can also run machine learning at the edge, speeding up the results of analytics and reducing the network load as only the relevant data is sent back to the cloud server.
By delivering faster and smarter analytics, In-Memory computing architectures can better forecast demand, proactively identify production line problems and eliminate them and create agile and responsive supply chains and enhance customer experience.
Smart factories are the future. When it is armed with sufficient speed, scale and performance – with intelligent access to the huge amounts of data, machine learning can introduce so many different types of efficiency boosters throughout the manufacturing organisation it will become an integral part of every manufacturer’s tool set.
Karen Krivaa is VP of Marketing at GigaSpaces. GigaSpaces provides the fastest in-memory computing platforms for real-time insight to action and extreme transactional processing. With GigaSpaces, enterprises can operationalise machine learning and transactional processing to gain real-time insights on their fast and historical data, and act upon them in the moment. The always-on platforms for mission-critical applications across cloud, on-premise or hybrid environments are leveraged by hundreds of Tier-1 and Fortune-listed organisations worldwide across financial services, insurance, retail, transportation, telecoms and healthcare.