Freedom to think
Top three applications for machine learning in manufacturing
Machine learning means machines do not have to be programmed to perform exact tasks on a repetitive basis, they collect data and use it to make informed decisions about their next move.
This allows them to correct any errors and improve their operational parameters. There are three key areas where manufacturers can benefit from this technology.
According to McKinsey, artificial intelligence can generate a ten per cent reduction in maintenance costs, up to a 20 per cent reduction in downtime and a 25 per cent reduction in inspection costs. Machine learning is a significant player in this positive impact of artificial intelligence.
In traditional predictive maintenance, engineers program the thresholds for a component’s normal operation into a supervisory control and data acquisition (SCADA) system. When the component deviates from normal operation, the system alerts an engineer of the developing fault.
The problem with this approach is the lack of flexibility. It does not take into consideration variations in plant activity or the context of manufacturing processes. For example, a system may detect a sudden increase in a component’s operating temperature and interpret this as a developing fault, when in fact it is due to the machine being sterilised.
Machine learning technology means predictive maintenance systems do not have to be programmed with normal operating thresholds. They use data from the factory floor and IT systems to monitor operational patterns and make informed decisions about what is normal and abnormal activity.
There are two main ways machine learning can improve quality assurance (QA). Firstly, it enables assembly robots to continuously monitor and optimise their processes.
Secondly, machine learning increases the capabilities of machine vision systems. Like with predictive maintenance, traditional machine vision systems for QA lack flexibility. For example, if a product is presented to a system in a lower illumination than usual, the system may interpret this as a quality defect.
Machine vision systems with machine learning capabilities use algorithms to optimise the camera and illumination settings for the object being inspected and for the environment it is operating in. They can also detect and localise objects without any operator input.
Collaborative robots work alongside humans but are only able to do this thanks to machine learning technology. Because the environment they work in is dynamic, they must be able to adapt to a large variety of circumstances, from things as simple as somebody blocking their route, to more complex situations like a new piece of equipment being introduced onto the factory floor.
This adaptability is important for ensuring the work is done quickly and to a high standard, as well as ensuring the safety of human staff. If robots perform the same actions repeatedly, regardless of their surrounding environment, they can cause injuries.
Siemens’ DexNet 2.0 robotic system demonstrates the value of machine learning capabilities in manufacturing facilities. Training a robot to pick up an object without dropping it requires complex programming.
The DexNet 2.0 uses a 3D sensor and machine learning to process information on the shape and appearance of an object and decide how to pick it up. As a result, it can pick up objects that it has never seen before.
Manufacturers should continue enabling human workers to have their own ideas and make their own decisions. However, they should also extend this liberty to their machines, to increase productivity, product quality and overall equipment effectiveness.
Luckily, you don’t need a state-of-the art system to introduce machine learning technology into your manufacturing plant. Older systems can be retrofitted with smart technology to help you make the most of the capabilities that this technology offers.