How to Use Big Data Without Creating Bad Data
Across many industries today, Big Data has become a core component of business decisions. It’s no longer just a buzzword—almost 70 percent of companies report they have been using Big Data for more than a year. But data is nothing without analytics. Data analytics can provide organizations with competitive advantages, including the performance insights required to lower costs and increase accountability.
Despite the proven value of Big Data, many industrial organizations are faced with messy, siloed data sets as they transition to fully digital operations. Large manufacturing and energy organizations collect megabytes of data across hundreds of thousands of instruments each day, providing ample opportunity for data analytics, but their limited data maturity results in mismanaged data and inaccurate insights.
When data is derived from a sizable number of physical assets and human operators in a plant environment, it’s often stored and organized in different systems, with different categorization. This is particularly critical when reviewing machine performance and system health. When characterizing an asset failure, for example, the first step is recording the failure mechanism and cause, followed by the maintenance performed to address it. In Computerized Maintenance Management systems (CMMS) like SAP, a field called “breakdown indicator” is used to record whether the asset failed, and a drop-down list is then available from which to choose the failure mechanism. If an operator is manually recording the issue and does not know how to define the asset failure or even the asset category, the field may be completed incorrectly or left blank.
In one account, a reliability team at an energy facility used raw data to try to determine how often their assets failed and to prioritize the fleet. Based on the raw data, the team calculated a turbine’s average time between failures to be 16,000 months. After looking at the data more closely, the team realized that the breakdown indicator field in their CMMS was rarely populated, so it was difficult to determine whether the asset actually failed. The team investigated the data further, and after reclassifying the data to accurately reflect failure, the average time between failures turned out to be just 14 months.
In the age of Big Data, it is critical that data be characterized correctly and consistently across all management systems to ensure organizations have an accurate, holistic view of machine and system performances. When one hour of unplanned downtime can cost a company more than $100,000 in lost production, optimizing asset performance must be a priority.
