The processes used to manufacture goods have changed beyond recognition in the past 200 years, through a series of steps characterised as four ‘industrial revolutions’. Alessandro Chimera, TIBCO’s director of digitalisation strategy explains.
First came the use of energy sources—water then steam—to power manufacturing processes, which required a consolidation around these energy sources. Then came the great efficiency gains from the streamlining of manufacturing processes—the assembly line—followed by automation and control, thanks to the development and deployment of digital technologies.
The latest revolution, industry 4.0, takes that one step further by turning the flow of digital information into a two-way street: data is used to control the manufacturing process, and is collected and analysed to improve efficiency and detect problems.
The challenge to anticipate malfunctions
There is one common denominator to all four phases. The operators of manufacturing systems and processes strive to anticipate and if possible, prevent any malfunction of production machinery that will compromise product quality or interrupt production. It’s not good to discover a machine has been producing malformed widgets after it has created a few hundred.
In industry 1.0 and 2.0, the ability to do this was limited and processes largely empirical: an engineer attuned to the sound of the machinery might detect a subtle change indicating a malfunction, for example. In Industry 3.0, monitoring many operational parameters enabled a one-to-one correlation between a malfunction and an alert: a light coming on to indicate a specific fault.
In Industry 4.0, masses of data are generated from all components of a machine and all stages of a process. Massive computing power is now available to employ machine learning and artificial intelligence techniques to detect anomalies long before they become apparent to a human observer or before they interrupt production or compromise product quality.
In today’s large-scale, high throughput, and extremely complex manufacturing processes, such techniques are becoming essential for the early detection of anomalies and performance degradations that can quickly have large-scale and costly impacts on the outputs of manufacturing processes.
Applying AI and ML
Real-time data from a machine or process can now be analysed to predict the likelihood of an anomaly occurring. Anticipating an event before it impacts product quality of operations can produce considerable savings in costs and time.
However, extracting maximum value from available data to detect anomalies early before they have a significant impact is challenging, even with the power of AI and ML. The datasets produced by manufacturing processes are massive and exist in multiple silos scattered across numerous systems.
So, the first stage in applying AI and ML to anomaly detection is to bring together data from these disparate sources.
The next stage is to employ technologies designed specifically to extract meaning from these disparate datasets, which can be achieved using an autoencoder. A good description of an autoencoder is “an unsupervised artificial neural network that learns how to efficiently compress and encode data then learns how to reconstruct the data back from the reduced encoded representation to a representation that is as close to the original input as possible.”
United States semiconductor manufacturer Hemlock Semiconductor (HSC) has successfully used anomaly detection to boost the efficiency of its semiconductor manufacturing.
Anomaly detection enables HSC employees to compare key parameters against pre-defined thresholds and optimal patterns developed by gathering a wide range of operating data and applying machine learning and AI.
As soon as the parameters for critical indicators for a process breach predetermined thresholds determined by this analysis an alert is raised enabling staff to determine exactly what has caused the alert and take remedial action, in many cases pre-empting any impact on production and product quality.
Over time the use of anomaly detection will lead to more energy and resource-efficient manufacturing and to more reliable and better-quality end products, all of which will be good for the planet.
In Australia, not-for-profits like the ARM (advanced robotics for manufacturing) Hub, are assisting manufacturers in embracing AI to overcome the present skills shortage. It provides expert advice on AI solutions and scientific and technical expertise. During its first year of operations, it reported it had worked with over 200 companies in this capacity, seeking ways to maintain and grow their digital capabilities.
Two successful instances are Verton Pty Ltd and Australian Droid and Robot (ADR) Pty Ltd are examples. With its world-first range of remote-controlled, load-management systems, Verton reinvented heavy-lifting operations, delivering “safer, faster and smarter crane operations”, but now with the addition of its AI capabilities, it is forming international joint ventures and looking for big data opportunities. This reach into AI was its response to the industry 4.0 call for digital capabilities to aid and improve manufacturing processes, particularly in the integration of advanced robotics and artificial intelligence.
ADR, on the other hand, does not lift cargo; instead, it drives and flies them. It has applied robotics and automation to industries including emergency search and rescue, where it developed a highly specialised drone and automated vehicles in mining.
Overall, the ARM Hub is delivering some exciting AI initiatives across many industries – from mass manufacture of construction solutions using platform technology, to the use of computer vision and ML algorithms to detect disease in ginger rootstock. It is increasingly successful in demonstrating how AI and machine learning technology may assist Australian manufacturers in increasing productivity and efficiencies.