IIoT for Predictive Asset Maintenance has generated significant interest in the wind farm industry. The following is a summary of topics covered in our recent article published in Power Magazine.
Why is IIoT for Predictive Maintenance so attractive to Wind Farms?
Up to 30% of the levelized cost per kWh produced over the lifetime of a turbine can be attributed to Operations and Maintenance. Unscheduled wind turbine downtime can last a week or more per year and in some cases, significantly more time. Because of remote locations, it is often expensive to transport heavy replacements. In the case of a generator, transportation can take more time than the actual repair. It is estimated that over 70% of wind turbine downtime is due to major repairs.
For operators, the high O&M cost is an opportunity for incremental revenue which is driving the interest in IIoT Predictive Maintenance.
What are the limitations of current maintenance techniques?
Predictive Maintenance (PdM) using SCADA data is the default system used today. How does it work? Sensor data from turbines are monitored based. If manually set control limits are breached, operators are alerted. There are two major limitations. First, only a limited amount of sensor data can be monitored. In many instances, the root cause of machine failure is from unknown sources. With traditional PdM, unless the root cause is due to one of the selected sensors, it will not be detected. The second limitation is based on how the SCADA data is monitored. If control thresholds are breached, alerts are generated. However, in many cases, once the control threshold has been breached it is already too late to prevent machine failure and remediate.
The lack of visibility into evolving downtime reduces the effectiveness of PdM.
What is meant by Unsupervised Machine Learning for Asset Management?
There are different methodologies used for IIoT Predictive Maintenance. Let’s start with the term “Supervised Machine Learning.” With Supervised Machine Learning, the learning algorithm is “trained” using human guidance and labels of abnormal and normal wind turbine conditions. In the case of a wind turbine, the learning algorithm needs to understand the physical layout of the machinery based on its blueprints. In addition, it needs training on mechanical processes. Once new data is analyzed, machine failure can be recognized based on the training of the algorithm.
This level of training is not required for Unsupervised Machine Learning. For instance, the algorithm does not need to be trained using knowledge of the mechanical processes of the wind turbine or its blueprints.
Automated Unsupervised Machine Learning is the next level of Machine Learning. In this case, models are self-maintaining and self-learning and can be applied to the various sensor and machine types.
What is the Economic Benefit of IIoT Predictive Maintenance?
The goal of IIoT Predictive Maintenance is to provide alerts of asset degradation and evolving failure early enough to prevent unscheduled downtime. If operators are alerted to failure, they can reduce workloads while parts are ordered. Reactive Maintenance is both expensive and more time-consuming. the need for reactive maintenance caused by unplanned machine breakdown.
The goal is lower O&M costs and higher yield rates.
Let’s look at some specific examples. The estimated O&M cost for Germany, the UK and Denmark are between 1.2 c€ and 1.5 c€ per kWh of wind power produced. It is estimated that approximately 50% of this is allocated to insurance, administrative costs, and other overhead costs. Using IIoT Predictive Maintenance, one can expect a reduction in replacement and labor costs.
Assuming a 20% reduction in the repair and maintenance port of O&M costs, this would translate into annual cost savings of $11,383 for a 2.5-MW turbine and $34,148 for a 7.5-MW turbine.
To read the full article, please click on this link.