Over the last few years, we’ve seen an increased emphasis on digital transformation by industrial operators. Large operational and machine health datasets have become a focal point to feed our optimization activities. Data scientists realized early on that machine learning analytics techniques were only as effective as the datasets from which they were trained. Good data drove convergence and clarity; bad data drove uncertainty and ambiguity. Good data was important.

Our focus on data was not without fault. We initially looked to big data insights – the results of our cutting-edge analytic techniques – as the primary contributor of value. We started to speak about control systems, SCADA and machine condition monitoring systems as data sources. We tried to convince ourselves we could ‘clean up’ the data in software. Our efforts fell short…

As it turns out, accurate machine health insights rely heavily on hardware, beginning with the sensor. Sensors perform a critical role in the creation of insights: they measure physical properties of a machine or process and convert these properties into signals that can be acquired by sophisticated measurement systems. Once filtered to reduce noise and further post-processed into data, they help create direct relationships to machine or process health.

Don’t be fooled by low-cost sensing options, which often sacrifice difficult-to-verify sensor characteristics, such as design reliability and measurement stability, in favor of cost-saving design and manufacturing techniques. The loss of value due to poor data quality quickly eclipses any savings of a low purchase price.

• Low-performance sensors often experience stronger effects from electrical interference emitted from machines or other electronic devices. This interference can manifest as an ‘apparent’ machine fault in the signal – and can be misinterpreted by typical software and analytics techniques.
• Low-quality sensors may experience measurement drift due to changes in the environment (such as changes in temperature) or the natural ageing of the sensors itself. These changes can make a mess of datasets, since they track the behavior of the sensor or environment, not the behavior of the machine.
• Low-cost sensors with short lifetimes can present as a nuisance in the plant. Even when the replacement sensor is warrantied, maintenance or instrumentation personnel have the time-consuming task of re-installing the sensor. Worse, faulty sensors often go unaddressed for months as higher-priority repairs or maintenance activities take precedence. During that time, vibration practitioners and analytic software are ‘blind’ to developing machine issues.

The true value of a sensor comes from the optimization or cost avoidance it delivers, like avoiding down-time and early fault detection, so that developing problems can be corrected at the lowest cost. Further value comes from not having to replace sensors during the life of a machine and through the creation of data not influenced by factors unrelated to the health of the machine.

The importance of data to digital transformation activities is inextricably linked to sensors delivering good data. It is my hope you will consider the entire signal chain, from sensor to monitoring system to analytics to insights to actions to outcomes. Every step in the chain is critical. Every step in the chain contributes to data integrity – or reduces it. A bad sensor compromises the effectiveness of the remaining steps in the measurement chain, invalidates insights, and leads to incorrect outcomes. For this reason, it is important to specify high-quality sensors and ensure physical measurements are accurately and reliability converted into the signals and data necessary for correct insights and successful outcomes.