A presentation was given by Bernie Cook, Director, Maintenance and Diagnostics, Central Engineering, Duke Energy during the ARC Forum session, “Moving to Predictive Maintenance with Industrial IoT.” This case story shows how the emerging technologies – IIoT and analytics – allow specific types of critical assets to have near-zero unplanned downtime while improving asset longevity and maintenance costs.
Bernie Cook presented the “Duke Energy SmartGen Program” which includes the application of IIoT for predictive maintenance. Duke Energy is the largest electric power holding company in the US with extensive fossil and hydropower operations in six states. It has four monitoring stations for reviewing the health of its power generation fleet. Business Driver
The primary reason for the new SmartGen program is to avoid catastrophic failures at power plants. In one case, Duke Energy had a transformer failure that cascaded into other transformers and two turbines, causing over $10 million in damages plus significant loss of power generation capacity and associated revenue.
An assessment of the cause of this incident pointed to the many manual data collection and analysis processes established over the preceding decades, in which meter readings, vibration measurements, and oil analyses were recorded on paper. In the case of the transformers, the readings and analysis were performed every six months. The paper documents were filed in cabinets spread across the five legacy companies that now make up Duke Energy. Unfortunately, an issue with an electrical bus accelerated a known minor transformer issue into catastrophic failure within that six-month inspection cycle. Solution
The significant financial loss drew management attention which, in turn, drove the review of condition monitoring and prompted initiation of the SmartGen program to better leverage technology to improve reliability and operations. To fill the time gap between inspections, engineering determined that online continuous monitoring was needed. This includes online sensors, a data management infrastructure, and equipment health and performance monitoring. Duke Energy built an advanced monitoring, predictive analytics, and diagnostics infrastructure providing a significant advancement in:
- Remote equipment monitoring
- Smart diagnostics & prognostics
- Data integration & visualization
- Enhanced reliability process (consistency across the company)
- Zero event operations (safety and environmental)
The new SmartGen infrastructure also provided a “force multiplier” to leverage the domain knowledge of a few specialists across the fleet of critical equipment. Their technical specialization and analysis improves reliability and operational performance.
For each type of plant, a model was built which helped to identify the sensors needed. The assessment included updating the failure modes and effects analysis (FMEA) for 10,000 assets in 50 of the more critical plants to identify the critical assets needing monitoring. Implementation occurred in three phases, with many of the easier items coming first and then moving to those requiring more resources. The monitoring and diagnostics system now has over 30,000 sensors, and uses the Schneider Electric Avantis PRiSM APR software for asset health monitoring and alert notification. PRiSM uses machine learning, which avoids the need to develop complex engineered algorithms, allowing Duke to build over 10,000 models. The system gives the company the visibility and decision support needed to be able to focus on the 10 or 20 things that need attention now out of tens of thousands of devices in the plants. Benefits
Mr. Cook gave an example of an issue that was identified early and avoided a $4.1 million expense. The monitoring and diagnostic center picked up small changes in vibration of around two mils after unit startup. The PRiSM software monitors patterns and notifies when small changes occur – well before people in operations are aware of the issue. In this case, PRiSM recognized a change in overall vibration information. Further investigation suggested that this rotor had a history of blade-to-shroud connection issues. A borescope inspection verified that several pieces of shrouding were missing. Since this was found during extremely cold weather, vibration levels were watched even closer for another change. The unit was taken off-line for repairs six weeks later.
New sensors, added data, and smarter analytics provide alerts that prevent the occurrence of costly equipment damage. A total of 384 finds during three years has conservatively avoided $31.5 million in repair costs. Duke Energy expects the rate of cost avoidance to increase further as it continues to train the machine learning models in PRiSM and adds newer sensor technologies.
Readers can view a video recording of Mr. Cook’s Forum presentation here