Predictive Manufacturing: The Future of Making

Predictive maintenance, using IoT to anticipate and prevent breakdowns by collecting and analyzing machine data, has been gaining momentum in recent years. Based on its successes, some innovators are applying the same kind of thinking to entire manufacturing operations and are even aiming to tie in visibility on the supply and demand sides.

In manufacturing, predictive an analytics generally refers to gathering data about a machine set ting or process and continually analyzing that data, explains Mark Wheeler, director of supply chain solutions at Zebra Technologies. “By monitoring the data on a regular basis, the manufacturer is in a position to correct an anomaly before it actually impacts product quality, yield rate, or some other critical outcome,” he adds. By identifying and tracking individual items, if an out-of-spec condition is detected, it can be isolated to a specific piece of equipment for attention and a specific set of parts for inspection and rework.

“Information volume is at the core of achieving system, process, equipment, or functional predictability,” says Saurabh Mehta, global head of markets, manufacturing, and logistics at Cognizant, an international business consultancy. Higher volumes of data make it possible to build predictive models based on longer and more dependable history. As an example, Mehta explains that the data related to product quality problems under different operating conditions can be immensely helpful in building models that can then be related back to the design and manufacturing process. Tracking back in this way enables manufacturers to predict potential quality catastrophes before they happen.

Define how to measure cybersecurity and then establish how much you need.
_{Jonathan Luse, CEO at Intel Industrial Solutions}

Data accuracy is directly proportional to the level of confidence one can place in the decision processes based on the models that are built. As an example, the accuracy of physical or work-in-progress inventory capture would make the decisions about line planning much more dependable. “In the process industry, this accuracy would help build models that would help minimize the off-spec or waste, thus improving overall process yield,” Mehta says.

Where Are Your Assets? Bluetooth smart beacons like these from Zebra Technologies offer real-time tracking and management of goods in complex logistical operations, including on-demand information about location and status of assets.

Based on things like existing root cause analysis practices, predict avoiding quality problems or major impacts and offer the ability to clearly define conditions under which quality problems most frequently occur. This enables manufacturers to clearly spec out their products and protect themselves. “A good example of this is a proper maintenance schedule that avoids significant damage for equipment whose components are known to fail after 5,000 hours of operation in defined ambient conditions,” he says.

“Overall, the advent of predictive capability from a quality perspective not only provides engineering teams with the ability to correct quality problems during design phases but also makes them completely aware of the conditional quality problems that cannot be resolved, thus making the designs more reliable,” Mehta adds.

Investment in predictive manufacturing may require some vision.
_{Saurabh Mehta, Global head of markets, Cognizan}

Putting Predictive to Work

An example of an organization on the path to predictive is Alpla, a manufacturer of plastic packaging for brands like Coca-Cola and Unilever. The company uses Crate’s IoT Data Platform to analyze data from tens of thousands of sensors. Processing is done in the cloud and a central control room monitors plant performance at local and remote facilities. From these insights, Alpla can identify trends at an earlier stage and its machine operators can be guided quickly to make necessary adjustments.

According to Mehta, the concepts of predictive manufacturing need to be applied in the contexts of the application or challenge, the larger business needs, and the extent of the potential impact. For example, the application of deep learning techniques to ensure a precision cutting process may be of less value than a simple measurement or control system. However, using deep learning is appropriate to predict quality variations given the complexity of the problem, as well as the magnitude of downstream impact. “Application of predictive manufacturing efforts without the context of the application and its larger business impact results in being either ineffective or over-effective,” he says. “The other important point to remember is to align the predictive manufacturing efforts to larger organizational digitization or transformation needs. Without that, the business benefits would not be proportionate to the effort and the traction would be lost.”

Teaching AI: One of the real strengths of machine learning is that there are different types of learning algorithms which can be used, including supervised, unsupervised, and reinforcement.

An Organizational Tune-up

Predictive capability can have a direct impact on productivity and can also help operators to work more efficiently. This may lead to the use of alternative materials in the design of a product – like composites, different types of lubricants based on conditions of use, or warranty parameters that provide heavy equipment fleet operators with guidance on how to manage their fleet and extend the life of key consumables. “Enterprises have traditionally been challenged with converting real time, historic OT [operational technology] data from legacy systems into higher-level IT insights,” notes Keith Higgins, VP of digital transformation at Rockwell Automation. Data produced on the factory floor needs to maintain its rich context (such as process conditions, time stamps, machine states, and other production states) to provide maximum insights to factory staff.

He notes that aggregating the data generated by machines in processes previously required significant manual effort and the pulling of information from many disparate sources. “By implementing advanced analytics software, including machine learning, within their manufacturing systems, organizations can automatically capture high-speed, contextualized OT data from industrial controllers in real time and generate predictive insights and operational excellence across their enterprise,” he says.

Data creates new opportunities and some complexity challenges as well. Jonathan Luse, general manager of Intel Industrial Solutions, says, “With emerging technologies like 5G and Wi-Fi 6, it will be easier and cheaper than ever to gather new data from your operations. Around my organization, we talk about implications of the ‘Three Vs’ of big data – volume, velocity, and variety.” He adds that Intel sees increased volumes of data being collected at increasing rates, coming from a variety of sensors (and linked systems). Luse notes that this can create both opportunities and new problems: “The ‘garbage in, garbage out’ concept holds true, and not all data is equally important, however it’s still important to gather as much data as possible to use it to dynamically discover actionable insights.”

The Artificial Intelligence of Things transforms raw data into business outcomes.
_{Bill Scudder, General manager for AIoT solutions at AspenTech}

With all that data and so many analytical activities, the rise of the Industrial Internet of Things (IIoT) is giving impetus to a new digital solution category – the Artificial Intelligence of Things (AIoT), says Bill Scudder, general manager for AIoT solutions at AspenTech. This new field is seeing the combination of AI with IIoT to enable the next generation of industrial AI infrastructure, allowing organizations to achieve more efficient IIoT operations and seamless human– machine workflows, to harmonize industrial data management, and the ability to transform raw data into tangible business outcomes rapidly, he says.

The concept of predictive manufacturing effectively extends an enterprise digital strategy. It should help reduce costs, increase quality and throughput, and prepare the organization to be more agile, says Naren Gopalkrishna, digital product manager at GE Digital. As the organizations mature in their predictive manufacturing journey, several other aspects of optimization are driven forward, such as predicted observations and prescriptive actionable insights.

IoT is making predictive manufacturing possible.
_{Mats Samuelsson, CTO at Triotos}

Organizations need to have a certain level of digital transformation maturity to successfully implement the predictive manufacturing concept. “The digital strategy should align with the larger manufacturing strategy and it should also consider the business problems that must be addressed,” says Gopalkrishna, adding that the IT and OT teams need to work together.

Predicting the Downsides

At Cognizant, Mehta’s view is that predictive manufacturing is a strong concept but implementation is often lacking in terms of providing sufficient volume, granularity, quality, and information accuracy. As an example, a temperature measurement at the output of a process can be effectively used to control quality in real time, avoiding quality issues by retrospectively analyzing the data. Lack of appropriate measurement (sensory) and/or intake frameworks would lead to the absence of this data, or the inability to use it even if it’s measured.

Zebra’s Wheeler says predictive manufacturing may require investment in visibility infrastructure to provide real-time data plant-wide. “Justifying this investment may require some level of vision of the broad uses and value of leveraging this visibility,” he adds.

Manufacturing in the Cloud: Embedded connected product monitoring enables data monitoring and analytics, administration, IoT device provisioning, and network operations through a single embedded board from a cloud platform.

Mats Samuelsson, CTO at Triotos, a company that builds overlay solutions on the Amazon Web Services (AWS) IoT cloud platform, sees the combination of better ways of collecting and processing data from new IoT technologies, plus improvements in machine learning, analytics, and AI, as a game changer. “They will certainly be combined with integration of existing and new control technologies for steady improvements in how manufacturing and production are planned and operated,” he says. “The question is which strategies enterprises will embrace to cost-effectively seize the opportunities, such as predictive manufacturing, that IoT is making possible,” he concludes.

READ MORE ARTICLES