Create a free account to continue

IIoT Value Problem: First-Time Data Costs the Most

The value of the Industrial Internet of Things (IIoT) does not lie in simple tasks such as operations measurements, but rather, in the vast amounts of data IIoT can collect, which can be used in more high-level and transformative...

The value of the Industrial Internet of Things (IIoT) does not lie in simple tasks such as operations measurements, but rather, in the vast amounts of data IIoT can collect, which can be used in more high-level and transformative areas such as research and development. Also of great importance to the new era of IIoT-based manufacturing is the contribution it makes towards the business goal of continuous improvement.

The value of IIoT is undeniable, and is fast becoming an essential part of any manufacturing operation that wishes to remain competitive. Yet, IIoT still has a significant value problem in industrial settings: The amount of data needed to collect and analyze a problem the first time is much greater than the amount of data needed to detect it a second time.

This is best illustrated in the following fictional example, which is based on real-world manufacturing problems and solutions.

In this example, a manufacturer has a Vertical Machining Center (VMC) that has a spindle with four ball bearings in it. If you instrument the machine with vibration, temperature, and sound sensors, and run it for many hours, you may identify some characteristics that indicate those bearings are going bad. Once you know those characteristics, instead of many sensors continuously monitoring and storing every signal, you can find the one signal to look for, and install a single sensor to look for it. Essentially, when you have a specific problem to solve, you can solve it much more cheaply if you know what you're looking for already.

Of course, this only applies to known solutions to known problems. A big promise of IoT has always been that it can help to identify unknown problems. Let’s use that same VMC as an example. You may evaluate that machine in operation, and determine that certain combinations of materials, bits, RPMs, coolants, and material removal rates (MRR) lead to chatter, poor surface finish, or decreased cutter life. The IoT can monitor the first category, chatter, but it probably can’t tell you about surface finish or cutter life, since the machine has not been instrumented to measure those issues. So, the machine operator needs to enter information about the chatter into some quantitative system. Then, if you find a relationship, you can compensate by modifying one or more of the input variables, all the way up to the CAM software level, and solve the problem by cutting differently, using a different coolant or a different coating on the bit, or changing the MRR.

Once you know this cause-and-effect relationship, you don’t need to monitor for it anymore. Instead, you decide to monitor continuously in case something unknown happens. For example, a cutter breaks when experience says it shouldn’t. But this unexpected outcome causes an exponential increase in the data you need to collect, because you already have data that says that the bit shouldn’t have broken. If your operator can quickly verify that nothing was obviously wrong (i.e., operator or programming error), then you would need to examine a huge pool of data to find a hitherto unknown signal that might have allowed you to take preventative action.

The problem is, it may not be worth the bother. To some extent, there may be a predictive signal you could discover, but it might not be worth the effort. There are very likely many “signals” in the world that are hidden in random noise, but the use cases where it is cost effective to extract them may be vanishingly small. There are also places where the volume of data needed to extract an accurate signal just does not exist or cannot be collected. For example, the relevant predictive signal could be outside of the sampling characteristics of the sensors — a temperature signal that is too short, a sound that is too quiet, or a harmonic oscillation that is too fast.

The cutters in vertical machining centers, or end mills, cost at most a few hundred dollars, and VMCs tend to manage production and safety well. The return on investment for this IoT data collection and analysis is small or non-existent. Does the value of IoT data collection and analysis apply perfectly to all use cases? Of course not. Jet engines may in fact warrant continuous investment in monitoring, data storage, and analysis. But, what are the diminishing returns for simple things? Motor bearings? Pump cavitation? Electrical shorts? The existing processes and data may already be adequately managing these risks for a given level of investment.

The value of IIoT goes much deeper than the underlying technology. Its contribution to research and development, and its usefulness as a means of achieving continuous improvement are where the real value lies. IIoT is built on data, but it is more than data itself — it is a combination of data, technology, capability, and perspectives and insights which can lead to operational improvements.

Critically, businesses looking into IoT should ask themselves whether they are looking for a one-time answer to a question, or whether they also need continuous monitoring. While the IoT is still nascent, one-off questions may be tempting to investigate, but following through on long-term monitoring and improvement can help align the costs of investments in instrumentation and analysis with the expected benefits of an IoT solution.

Alex Bakker is Research Director at ISG Insights, a unit of Information Services Group (ISG)

        An unexpected error occurred: Timed out after 10000ms
More in Industry 4.0