Moving Beyond the Basics: Machine Learning and AM
25 April 2018
In the era of Industry 4.0, manufacturing is increasingly heading towards a world of machine learning and artificial intelligence. A world where data-driven systems can be developed to enhance production processes. And additive manufacturing can leverage the benefits of machine learning to achieve greater efficiency, enhance product quality and optimise AM workflows.
Increasing efficiency with machine learning
As additive manufacturing scales up for end use production, advances in machine learning are not just limited to the prospect of self-driving cars. Machine learning can be used within additive manufacturing to drive efficiency in part by eliminating trial and error methods during the production process.
A large number of factors, such as part orientation or the design of support structures, can potentially affect the material structure of a part and lead to build failure. This inevitably means that the reason behind the failure of a build can be attributed to a number of variables. Typically, a trial and error approach has been applied in order to achieve a reliable printing process. However, as this involves going through a number of failures before reaching the optimal process,a trial and error approach inevitably lacks efficiency. Machine learning can help bypass a trial and error approach to production by developing a system to help machines determine the variables and parameters in advance, thereby optimising the production process.
The U.S. Navy’s Office of Naval Research (ONR) has recently partnered with data company Senvol to develop machine learning software that can analyse the relationship between AM process parameters and material performance. The aim is to enable ONR to reduce reliance on traditional material testing.
And research by the ADAPT Center in Colorado has already begun to explore how machine learning can identify the inside geometry of the part, predict the correct parameters for any new part and therefore optimise the printing process.
Using machine learning to enhance quality processes
Machine learning can also be implemented to add another layer of quality control to the production process as machines are eventually able to self-correct and supervise themselves. Using machine learning technology, large amounts of data can be analysed and used to provide a real time status of each production stage. Machines can use algorithms to find patterns in production data, and from this construct predictive models, refined through comparisons with real data.
Last year, GE revealed its research into the use of machine intelligence and digital twin to enhance machine and material performance for metal 3D printing. Through its research into machine learning, GE aims to reduce material waste through the detection of quality process problems, with the ultimate goal being a 100% yield. GE’s research aims to achieve full visibility into each layer of a part build, training the machine to recognise problems with the build itself. This will allow users to see the mechanics and structure of a build as well as identify issues earlier in the process.
Other use cases of machine learning
Additive manufacturing has proven to be an ideal solution for the spare parts industry, due to the high costs of storing and maintaining an inventory of spare parts. Additive manufacturing solves this problem by enabling manufacturers to produce and supply spare parts on demand at the point of need.
And yet, machine learning can take this solution one step further to improve the efficiency of the production process as improve predictive capabilities. In the case of discrete manufacturing for example, companies can make use of predictive maintenance models to predict the lifespan of a specific part. Machine learning can also be used to identify when a customer needs to replace parts using a preset data schedule, enabling manufacturers to send replacement parts ahead of time. Manufacturers should therefore consider using machine learning to reduce costs and ensure higher customer satisfaction.
Machine learning – a huge potential for AM
Machine learning has the potential to enhance production processes, guide decision-making and ultimately transform business models. Applications of machine learning for AM are numerous, ranging from enhancing design processes to improving efficiency and even determining the printability of a 3D object before the printing process has begun. However, implementing machine learning and AI systems is also not without its own challenges, requiring strategic planning and investment in both software and hardware infrastructure. But within the era of Industry 4.0, it is clear that the use of machine learning, AI and big data for AM is only at the tip of the iceberg.