Key technologies defining smarter factories: AI
Artificial intelligence (AI) is an area in which activity is now ramping up at a phenomenal rate. There are many different constituent branches to AI – the more popular ones include data mining to find previously unknown relationships between data points, and machine learning (ML), this can help create smarter factories.
By Mark Patrick, Mouser Electronics
ML is a crucial component of Industry 4.0 and will help define how the smart factories of tomorrow operate, and in this instalment of this six-part blog series we will be looking at it in detail.
With ML, data analysis can be automated (based on established models). Automating data analysis requires that the algorithm applied first learns to identify known patterns. The machine learns from a framework (or model) of historical datasets, where the relationships between data points are clearly labeled. Once it knows how to spot known patterns, the machine can analyse the data collected from sensors to make predictions based on the learned parameters. This process is called inference.
In a manufacturing setting, ML training can employ historical data from the machine being taught, or if such data is not available, historical data taken from comparable machines. A trained ML algorithm will analyse the data collected by sensors installed on different machines; identify patterns, such as images, audio (speech or sound) or other metrics; and then predict outcomes.
ML can make factories ‘smarter’ by improving quality assurance, automating processes and increasing production output, supplanting the pattern recognition mechanisms previously used (such as Kalman filters), where acquired imaging data was just directly compared to a reference image (as there was no ML inferencing involved). It can also be pivotal in ensuring worker safety and security.
Improving quality assurance
ML software’s capacity for image recognition enables the identification of defective products. While humans will get fatigued or bored with repetitive work and start making mistakes or overlooking things, robots will not. In low-mix manufacturing environments, robots are perfect for work that is high volume, repetitive and follows fixed procedures. In high-mix environments, robots can be trained through use of ML algorithms to grasp, rotate and inspect objects properly. By using high-resolution cameras and sensors for image analysis, the robots can check for dimensions, color and other required characteristics.
For example, one system (DeepEyes) utilises 1080p resolution cameras, which can detect three-dimensional anomalies as small as 0.01mm from 1m away – even if the products are on a moving conveyor belt. ML-trained robots can also verify that the product contains the correct number of parts, check for proper installation of those parts, and find product defects (such as scratches or fractures). Lastly, they can make sure that the various parts are combined correctly for specific customization tasks.
McKinsey estimates that across more than 800 occupations, more than 2,000 work activities have the potential to be automated through application of AI. Due to their ability to perform highly repetitive work quickly, robots are the perfect conduit for automation. Collaborative robots (cobots) – as discussed in our earlier blog – can work alongside humans and complement their work.
These can also be trained via AI to perform more nuanced tasks, such as pouring a small amount of polymer liquid into lens molds without creating bubbles, or welding parts of a product together. In addition, AI can help them navigate the factory floor on their own.
Adding to operational efficiency
AI can boost operational efficiency on multiple fronts, including streamlining supply chain logistics, reducing delays and raising work environment safety levels. With consumers demanding more product customization, enormous pressure is being placed on manufacturers to adjust production systems accordingly. Via ML, it is possible to leverage data analytics to improve the supply chain, reduce the cost of raw materials, increase communication between teams and diagnose problems faster.
Also, by analyzing the flow of raw material and the movements of personnel across the factory floor, AI can help reduce congestion, avoid collisions and decrease the amount of non-value-added work.
In addition to increasing output, AI can help reduce the number of production delays by preventing the occurrence of hazardous events and removing personnel who may pose a potential liability. Traditionally, facilities have had a designated team of workers patrolling the floor to identify and rectify dangerous conditions. Such patrols are slow, infrequent, inefficient, labor-intensive and error-prone.
However, AI can analyze sensor data and pinpoint potential hazards – such as personnel entering forbidden areas while machinery is in operation, misplaced combustible materials, or an area overheating to a dangerous level. Employees or contractors who are not in an appropriate physical state for work because they are too tired, intoxicated or distressed, for example, can be identified with AI. Furthermore, by using AI it is possible to make sure everybody is wearing appropriate safety gear (like helmets, goggles, work boots, hearing protection, etc.).
Ensuring the security of a factory is also crucial. Theft, burglary and industrial espionage all incur high costs in the form of lost inventory, delayed production and compromised business intelligence. AI-enabled recognition of patterns in faces, fingerprints, iris scans and voices helps to control access to sensitive or restricted areas. Even in the absence of biometric data, AI surveillance can identify an unusual traffic pattern, as well as the presence of inappropriate personnel outside of predicted schedules or in the restricted areas, and subsequently alert management.
Over time, an AI algorithm can continue to improve the accuracy of its predictions by learning from the new data collected by the sensors around the factory. With access to the latest data, data mining software can identify previously unknown associations between parameters, which may lead to new or different kinds of process improvement.
In the future, there will be more combined use of technologies. The fields of augmented reality (AR) and virtual reality (VR) are expanding quickly. Combining AR or VR with AI will lead to better-simulated environments for product development and improvement, as well as ongoing production planning.
Much of this blog series has been concerned with data – its acquisition, transportation, analysis and manipulation. In the final part, which will be published shortly, we will focus on what needs to be done to protect it from malicious parties.
What will this blog series cover?
- Key technologies defining smarter factories – connectivity
- Key technologies defining smarter factories – sensors
- Key technologies defining smarter factories – the rise of cobots
- Key technologies defining smarter factories – digital twinning
- Key technologies defining smarter factories – AI
- Key technologies defining smarter factories – data security