The food and beverage industry has been shifting to adopt more artificial intelligence (AI) applications to improve and address issues of operational efficiency, food quality and safety, sustainability, and supply chain challenges. In a recent webinar, Demonstrating the Power of AI for Food and Beverage Innovation, Sridhar Sudarsan, chief technology officer at SparkCognition—a provider of AI systems for the industrial space, including manufacturing—discussed how AI and machine learning (ML) can be invaluable tools in addressing these issues.
Sudarsan stated that over the last year and half, companies have had to learn that demand is always evolving, often resulting in challenges faced on the manufacturing side with inventory, line optimization, maximizing production, minimizing unscheduled downtown, and increasing overall efficiency. Additionally, according to the 2021 State of Manufacturing Report, by Fictiv and Dimensional Research, quoted in the webinar, 31% of manufacturers are concerned about lack of visibility throughout the supply chain.
To explain how AI addresses these challenges and concerns, Sudarsan compared the supply chain to a universe with the manufacturing plant at its core. He included upstream and downstream supply chains to close the loop. Both the supply chain and the plant involve thousands of different types of systems, hundreds of processes, numerous assets, and hundreds of people that interface with the production of materials and products daily—and the essential tie between them is data.
Sudarsan added that “It isn’t enough to have the data being produced and used. It’s about making the data consumable and usable in the right way. It’s about making the data actionable.” Sudarsan broke down the process of making data actionable into four steps along with considerations companies would take within the steps, which are data availability, leveraging data, usability of insights, and scaling infrastructure.
1. Available data vs. actionable data
Data sets come from a variety of machines, sensors, logs, maintenance reports, shipping reports, image data from cameras, physical documentation, worker tribal knowledge, and more. Oftentimes, a company has dark data already at its fingertips. Yet, the raw data that can be gathered from these sets, though available data, is not necessarily usable. In addition to the usability, collecting data requires monetary investment. As Sudarsan said, “It’s not enough to collect a lot of data and have that sitting somewhere in a storage system. It is important to collect the right set of data.”
In order to be efficient in terms of cost and collection, Sudarsan suggests an iterative approach, which does not necessarily require upgrading or changing an entire line. A company should be able to retrofit to find the right data to collect. And though a company may be tempted to spend significantly to research what the right data is, Sudarsan’s approach cautions against this. Determine key problems in quick iteration, as key areas can often turn out to be completely different from the research’s findings as problems and needs are uncovered. But keep in mind a wider, long-term goal that includes better current management of data and future proofing capabilities to avoid wasting money and resources.
|Vegetable Ingredients Supplier Goes Digital|
2. Insight generation to leverage collected data
The next step is to cleanse the collected data, making it usable to generate insights. This allows operators, supervisors, plant managers, regional managers, etc. to take the right set of actions at the right time. A company must build and use AI and other models to process the data that was aggregated to then be able to generate those needed insights. The model a company chooses must be able to process rapidly and find patterns, determine anomalies and deviations from the normal—which, as Sudarsan pointed out, changes continuously and can be monitored using normal behavior modeling (NBM), which determines the risk of the system at a particular time.
3. Usability of Insights
Insights, like usable data, are still useless, however, if they are not made available to the right people or systems at the right time. The most efficient way to ensure that availability is to integrate the AI model into a company’s ERP, MES, or CMMS system. “Integrate so you’re not trying to rip and replace every single system you have,” said Sudarsan.
Whatever system a company chooses, Sudarsan emphasizes that it needs to be a learning system. “Just as we continuously ask our people to up-skill themselves, to learn and adapt to things, we also want our systems to do the same,” stated Sudarsan. “A learning system is key so that as your processes, machineries, etc. drift, your models continue to evolve as well.”
4. Scale infrastructure
Once the AI model has been tested for a plant, or a plant in one area experiences insight useful to all company locations and products, rolling those changes out across the board is the next step to continuing the digitization journey. Sudarsan says this can be summed up as a “virtuous cycle” or continuous cycle, which consists of these four steps being taken in “bite-sized chunks” as a single unit. This applies to both larger companies and those aspiring to grow their footprint. As Sudarsan emphasized, it’s not about seeing value in three years or three months, but seeing immediate benefits through implementation of the virtuous cycle approach.