Countries take holidays at different times of the year, and the whole world can face downtime during wars and pandemics. However, outcomes from screeching halts can inform future production to make impactful use of IoT data and systems.

“Golden Week” refers to a collection of consecutive national holidays in Japan and China that creates extended time off from work. (This is not to be confused with Ohio’s former early voting period allowing residents to register and vote on the same day.) During Japan and China’s Golden Week, employees enjoy a long break while manufacturers risk breaking the bank without adequate planning.

Companies worldwide remember the pangs of prolonged downtime during COVID, and many chose to adopt more automation for lights-out production. Such problems persist as manufacturers continue facing worker shortages and supply chain disruptions.

However, challenge breeds innovation, and there are always lessons to be learned. Arjun Chandar, founder and CEO of IndustrialML, knows this well.

“If you don’t have production on the weekends, and you don’t have an automated system, then it’s lights out on Friday. First thing Monday morning, you don’t want to waste time ramping up — you want everything done already,” said Chandar.

The same goes for Golden Week or any planned or unplanned period when staff are scarce.

“This year, Japan’s Golden Week holiday was characterized by a significant influx of travelers and the closure of various businesses due to the easing of COVID-19 measures, posing challenges for manufacturers. The resulting disruption from Golden Week often leads to delays in production, hampered logistics, and decreased efficiency,” said Chandar. “Additionally, ongoing pandemic-related issues, such as high cargo demand, staff shortages, port closures, and limited shipping containers, compound the delays caused by the holiday.”

So, how can manufacturers, especially those with IoT ecosystems, refine operations to smooth out the peaks and valleys?

Use IoT to support the workforce

Machine learning (ML) extends IoT capabilities beyond remote monitoring and control, providing insightful analytics, anomaly detection, real-time optimization, and more. But operations include more than machines — there’s a people component that’s trending downward as experienced workers retire and post-pandemic shortages remain unreplenished.

“Many people in manufacturing are retiring, and younger people operate under a different paradigm where they don’t stay in jobs for long. So, with higher turnover, you’re losing a lot of domain knowledge,” said Chandar.

IndustrialML integrates data from various sources, including machine sensors, software, video, and audio. Users interact with the information primarily through three modules: real-time data visualization, no-code historical reporting, and a real-time communication system.

“Whenever something important happens inside or outside of a factory, such as when an order changes or a furnace temperature is too high, we can send a real-time alert to people with instructions on what to do,” said Chandar.

The real-time communication system can simultaneously send alerts from the same event to multiple people via different methods. For example, the system can send a message to a manager via Slack, an operator via a headset, and a salesperson via email to expect a delay or indicate an issue.

“That’s where the value proposition of real-time communication comes into play,” said Chandar. “Not only do we have a way to identify the issue, but we also can impart the necessary steps to the right people at the right time so they can address the issue without needing 10 years of experience at the factory or memorizing the steps.”

Chandar admits that the system doesn’t do anything new from a technical perspective but dramatically influences how integrated technology can positively impact the workforce.

“You can’t lose 100% of your expertise and still operate the factory. Domain knowledge has to come from somewhere, and somebody needs to maintain the content of the alert policies,” said Chandar. “But my ideal goal is that somebody can run a factory 90% as well after half a year as they would have previously done after five years.”

Get disparate systems talking

IndustrialML collects data from a factory’s manufacturing execution system (MES), which tracks each order produced on the floor. The MES system sends the order information and all corresponding nominal settings. For example, data for a steel tube might include the outer diameter, thickness, length, and coating.

Different products may have differing requirements, so optimal machine settings change based on the product in production. For example, the ideal temperature might be 450° F for one product and 430° F for another. Inefficiencies or false alerts can arise if the machines and systems don’t communicate. To prevent discrepancies and make actionable use of disparate data, IndustrialML’s alert policy system can perform comparative alerts to send the right information.

“You don’t want to set an alert policy that notifies you when something goes above 450° because that might not apply to everything,” said Chandar. “But if you can set a policy to send an alert whenever the temperature is 15° higher than its planned value, then you’re synthesizing the actual value from the sensor with the planned value from the ERP system. Data from disparate datasets can be compared against each other in one setting, and you can legitimately offer real-time communication regardless of what product is being made.”

Systems filled with untapped data limit operations and burden digital resources. But when they talk to each other effectively, enhanced capabilities abound and unlock more optimization potential.

For example, a real-time video stream of production parts can be captured and analyzed with an ML algorithm to determine the probability of a defect. With IndustrialML, manufacturers can create a policy to send an alert if the defect probability for any part is greater than 80% — or greater than 80% for more than 75% of the time, during the last 20 seconds, or other tolerances. Though the system evaluates all parts, only parts that violate the alert policy need to be isolated and inspected.

“Say you want to inspect 5% of your parts. This takes away that need and instead inspects 100% so you can target parts instead of taking a random sample,” said Chandar. “And because we’re collecting all the other data around that, if you have a defect, you can figure out the settings that led to it. So, on top of the communication infrastructure, machine learning allows you to learn new things, and collecting all the data together allows you to improve your process for the future.”

Optimize production planning

Although some shutdowns, such as those during Golden Week, are expected and recurring, they require effective planning to recover quickly and prevent unnecessary delays. Chandar advises factories to put mechanisms in place to track additional demand during downtime, plan production to meet demand or forecasted demand before the shutdown, and schedule at least the first few days of post-shutdown production before turning off the lights. Manufacturers can mitigate disruptions and streamline workflow by anticipating the impact on supply chain operations and adjusting production accordingly.

“One thing you can do if you’re collecting information from an ERP system is ensure that all the right work instructions are set up and everything is programmed so that operators have the steps they need immediately upon returning,” said Chandar. “Once you turn the factory on, operators will get the alerts and be able to execute on them. That will make it faster to get up and running than if you had everything on pen and paper, and operators needed to be told exactly what to do and perform setup from scratch.”

For instance, manufacturers may need to factor in long times to restore equipment to normal operating states. Instead of shutting off machines and dropping temperatures dramatically, they can keep equipment at moderate temperatures by modifying an alert policy with the desired settings. Then, they remotely monitor and control the equipment and program it to ramp up before operations resume — whether after Golden Week or a typical weekend. This also helps with energy consumption and overall sustainability efforts.

“In the long term, my vision is that something like this could operate up and down an entire value chain,” said Chandar. “For example, steel has become way more expensive, so people are looking to source it from different places, which introduces a range of potential quality issues. If suppliers are required to submit information, automatic information sharing between suppliers and customers could make supplier quality much easier to measure when trying to diversify your supply chain.

“In the other direction, if you’re selling steel tubes with embedded sensors, you can track all the information on the product and monitor its performance in the field. If you identify an issue, you can contact all your customers with an upgrade or explain the settings under which the product might fail. That’s extremely valuable information to have.”

Until then, IndustrialML is primarily used to help inexperienced operators perform work tasks effectively and help engineers and managers evaluate equipment performance and improve productivity. The intention is to make factories smarter by integrating disparate data and giving people the right knowledge precisely when they need it.

IndustrialML
industrialml.com