The Internet of Things (IoT) presents an opportunity to collect real-time information about every physical operation of a business. From the temperature of equipment to the performance of a fleet of wind turbines, IoT sensors can deliver this information in real time. There is tremendous opportunity for those businesses that can convert raw IoT data into business insights, and the key to doing so lies within effective data analytics.
Characteristics of IoT Data and Analytics
Innovations such as in-memory computing and massively distributed processing frameworks (like Hadoop) are important IoT catalysts for organizations that don’t want to compromise speed when analyzing millions of data points. As these technologies have progressed in tandem with reduced sensor and wireless network connectivity costs, IoT analytics initiatives have approached a threshold point making investments more accessible to a broader population of organizations. The IoT data that organizations collect comes from machine sensors and logs that are not part of the traditional corporate data analysis lexicon. As such IoT data feeds tend to have a number of unique characteristics that must be accounted for. Given the persistent nature of IoT data feeds, successful streaming analytics is at the core of fundamentally unlocking value from connected devices. Table 2 details the most salient characteristics of IoT data.
Choosing the Right Analytics Solution
In each observed instance, value to the organization – whether from cost savings or new revenue opportunities – was unlocked from the analysis of the collected data. That is to say, the process through which individual signals from sensors becomes actionable is through analytics. Broadly, Blue Hill observes four basic stages that represent the maturity of IoT analytics projects. These stages range from replacing manual data collection efforts with sensors to automatically making changes to operations without any human intervention. Each stage builds off of the previous one, and provides tradeoffs in terms of their complexity and their potential value to the organization.
Descriptive Analysis Descriptive analysis offers an opportunity for organizations to understand the state of affairs of their operations. Whether through data discovery efforts or building dashboards, descriptive analysis allows business decision makers to drill into specific areas of the business to identify performance levels, anomalies, and root causes of top-line outcomes. IoT initiatives present a distinction as this analysis can be handled both on ‘at-rest’ data and live ‘in-stream’ data. With IoT sensors, organizations have the opportunity to monitor key metrics and performance numbers in real-time by bringing the live data feeds directly into their analysis. This allows organizations to have the most up-to-date understanding of operations, and ensures that decisions are being made on the most relevant data points. However, not all IoT analytics initiatives rely on real-time decision-making. Even if the analytics systems bring in live streams of data from the field, organizations may prefer to review the data in aggregate at the end of specified time periods. As an example, the California based oil & gas company is able to monitor the performance of oil wells at the end of every day or week. This allows them to identify opportunities for improvement (such as increasing production levels) and areas of potential concern. Ultimately, they are able to take this information and disseminate it to their field crew to make adjustments or repairs. The result is reduced downtime and increased production levels. The company estimates that they lose $500 for every hour that a single oil well is not in operation. After analyzing the initial impacts of sensor deployment they estimate that quicker oil well repairs saves approximately $145,000 in cost avoidance per month per field.
Predictive Analysis
Using available data to create forecasts and predict future outcomes presents a next step for extending the value proposition of analytics efforts. In this regard, Blue Hill observes important implications across a spectrum of outcomes, from increasing asset uptime to strategic decision-making. For example, a manufacturer monitoring their production line can gain significant advantages by forecasting likely equipment failures before they happen. They may have sensors that monitor characteristics such as temperature, uptime, and output levels of a key piece of equipment. Predictive models built from these monitored sensor readings, combined with historical performance data, can identify leading indicators to future breakdowns. Making adjustments or repairs during scheduled downtime is far more cost-effective than disrupting planned production cycles. The result from predictive analysis in this sense is a reduced number of breakdowns and a higher level of production. This, in turn, may mean additional revenue as well as lower costs. Predictive modeling and machine learning algorithms are built and trained upon past observations, and must be continually refined based on realized outcomes. Again, IoT initiatives presents a distinction in that predictive analysis can be performed based on already-stored data at rest and with live in-stream data. Analytics environments that can stream data live enable dynamic forecasts to be delivered to critical decision makers in real time. This affords the potential for substantial competitive advantages and efficiency gains by reducing response times and generating more accurate predictions. Predictive analytics requires both an understanding of what statistical insights are possible and the tools to create accurate projections. As a starting point, this means avoiding basic statistical fallacies associated with error, interpolation, extrapolation, and accurate regression modeling to build a portfolio of outcomes. This can also mean using predictive analytic tools not only for operational data, but also for relevant text inputs, work schedules, and other external data that can provide context to the prediction at hand. Predictive analytics also require the same clean data needed to conduct historical analysis as a starting point to accurately benchmark and project success.
0 Comments