Maintenance is a significant cost factor in many organizations, and is under constant pressure of cost reduction. Unfortunately, there is also a tendency to only highlight the costs and disregard the benefits of a good maintenance policy. This could be due to the difficulties in rating and estimating the contribution of maintenance to the organization’s efficiency. This is exactly where big data and IoT – “internet of things”, can clarify matters significantly. How is that? Let us examine it in more detail.
A recent Forbes article described how GE uses some of its jet engines to deliver utility power and then sells power by the hour. When one of these jet turbines breakdown, it can be costly affair – just one hour of stoppage can result in lost revenues of more than $2 million. By using sensors and machine to machine communication – which is what the internet of things really implies – GE can detect and even predict failures before they happen and thus save significant costs.
This clearly has important productivity implications. If an equipment is experiencing issues, the traditional way to handle this would be going through the human information chain: a line worker notifies a supervisor who in turn verifies the issue and calls for expert maintenance crew. All of this can add cost and cause disruptions. With machine to machine communications enable by internet of things, we change this problem scenario. Now the equipment supplier or the so-called original equipment manufacturer (OEM) has real time awareness of the state of health of every equipment they have built and installed at their customer sites, and can detect these issues much faster or in some cases before they happen!
How can we detect faults or failure before the problems happen? This is where maintenance using predictive analytics comes in. Data mining concepts and methods can be employed in several ways for preventive maintenance. Production data, machine functional data, and sensor data are all aggregated for analysis and used to build models for predicting machine failure or poor product quality reducing failure times and costs. This process is in principle, no different than the ones currently used in fields as diverse as fraud detection in financial services or customer retention in marketing.
However handling this volume of data requires an engine that can integrate all the elements of the technology stack, extraction-transformation-loading (ETL), data cleansing, data storage, reporting, statistical modeling and data mining into either a single platform or a seamless stack, that can scale up to 10s of terabytes, handles multiple large tables of billions of rows, and executes 10-100x faster than conventional relational database technology. This is the domain of big data and IoT.
The bottomline to establishing a solid and reliable maintenance policy is paying close attention to a measure called “Availability”. Availability may be defined as the ratio of equipment delivered (or made available) on time to the total number of equipment requests. Increasing availability by either detecting failures faster or anticipating failures through predictive analytics can improve all of the following traditional key performance indicators (KPI) for maintenance, which are :
- Mean time between failures (MTBF)
- Mean time to repair (MTTR)
- Total cost of ownership (TCO)
Since availability has a substantial impact on the efficiency of a system, sophisticated and state of the art maintenance strategy optimizations will also need to use models that can accurately capture, measure and predict availability. There are several well-understood methods that can be applied to model availability, for example, Petri-Nets or Markov-Chains. However, in reality, practical maintenance strategy optimization approaches very rarely make use of these quantitative models. There are many reasons for this, but one important factor is inability to quantify the impact of preventive maintenance on availability. But big data and IoT with predictive analytics has the potential to change this for good.
Originally posted on Wed, Oct 16, 2013 @ 07:32 AM