Industry Week says IoT is a game changer because it provides component level instead of batch level visibility. That’s one use for IoT on the factory floor. The other is to use IoT to gather big data and then run analytics over that to improve planning, operations, and the efficiency of the supply chain.
Here we explain how manufacturers can use IoT to do both and why doing that is important for factory optimization.
Factory Optimization: The Long View
Optimizing factory operations is by nature a mathematical problem. So it fits perfectly into the IoT and big data analytics model, which is designed to gather quantifiable metrics to provide insights and make protections.
The mathematician would say that setting the optimal level to run the factory is a linear programming problem. That determines the optimal levels of inputs and outputs given certain constraints. That perfectly describes the assembly line and production planning, both of which are governed by capacity constraints and the speed and timing by which components become available.
Since early time and motion studies, thinkers of all types have tackled this problem of how to optimize the factory. The most famous of these was the American W. Edwards Deming. Douglas MacArthur sent this unknown statistician to Japan to rebuild their industry from the ashes of war. He literally taught Toyota how to build cars. Now manufacturers around the world study The Toyota Way.
The next group of thinkers to look at how to fine tune factory operations and the assembly line were what we used to call operation researchers. But their impact was not significant. Their correlation, classification, and machine learning models were so complicated that no one but themselves could understand them.
Enter Big Data and IoT
But all of that has changed now because IoT, big data, and analytics have made it possible to use applied mathematics on the factory floor without needing to understand the details of the logic behind the algorithms.
In the past few years, new big data databases, like Apache Spark, have emerged to let businesses process streaming data in a way that the old-fashioned, rigid data warehouse could not. And new programming SDKs have made it possible for ordinary programmers to apply operations research data models and machine learning algorithms to the optimization problem. IoT clouds and SDKs transmit data to these databases and algorithms via network-connect sensors attached to components and machines.
Tracking the Assembly Line
There are an infinite number of ways to tap into the real-time flow of information coming from the factory floor, warehouse, and supply chain. Yet, manufacturers have been somewhat slow to embrace the digital model. That means there is much room for improvement. However, when the manufacturer decides to move in that direction, typically the first item they focus on is preventive maintenance. Logistics often follows that. Those are the easiest items to optimize, because there are so many uses cases to follow.
General Electric is one company that has embraced IoT. Not only does General Electric have 10,000 IoT sensors at their Duracell plant in Schenectady, N.Y., they stick a sensor in every single battery.
The IoT-equipped batteries radio their location as the move across the factory floor and through the supply chain. This gets rid of the manual barcode scanner. And to keep the process of manufacturing batteries on track, sensors measure weight, volume, ambient temperature, vibration, noise, pollution, and light levels in machinery and components throughout the manufacturing process. Variations in any of those signal an operation that has gone out of bounds.
The Simplicity of Going Digital
IoT vendors and analytics cloud companies—like Google Cloud Prediction, IBM Watson, and DataBricks—and articles in Forbes are quick to say that IoT and analytics will “speed time-to-market, reduce TCO, improve customer relations, etc.” But what is often missing from the vendor landing page and the journalist’s article is exactly how to do that.
The technology and tools have become cheap, easy to deploy, and easy to understand.
The sensors themselves often cost only about $1. An Intel Edison computing card costs $35. And there is simplicity and low cost in the network too as there is no need to send out a network engineer to bring those online. ZigBee, Z-Wave, and BLE (Bluetooth 4) transmission protocols can work in peer-to-peer fashion. That means only one end of the network needs an IP address and some kind of controller.
Some companies have even started to attach sensors to human beings to track the efficiency of their motion and where they linger. Of course not every employee would be in favor of that.
Then there is the IoT cloud, which manages the IoT devices. That pushes out software updates and instructions and lets the operator reboot those as needed.
The Planner’s Dashboard
Data gathered like this from the factory floor, warehouse, and supply chain flows into an easy-to-read dashboard. That lets planners, managers, and assembly line workers visually track what is happening right now and make changes accordingly. That approach is several orders of magnitude better than looking at metrics gathered after the fact—which is what companies did before IoT and big data. The modern approach eliminates the lag time between finding the problem and fixing it.
With regards to the entire company, digitizing the factory floor is only half of the equation.
The company-wide analytics model adds transactional data coming from the ERP and CRM systems, consumer buying pattern data purchased from data brokers, and brand perception on Twitter and feeds those into predictive algorithms and machine learning models. That lets tactical and strategic planners react to changes in forecasted demand, currency fluctuations, what the competition is doing, and set prices, change the product mix, and set output accordingly.