Reading Time: 13 minutes
Operational and Maintenance (O&M) costs of the PV plants can have a big impact on its viability and success. Preventive maintenance is a traditional technique which involves periodic scheduling and reactive maintenance involves identification and correction of unexpected deviation of performance in the system. These traditional techniques take much time and are expensive not only in terms of manpower effort and money but also costs associated with downtime.
We can reduce Operations and Maintenance cost by digitizing the power plants, that collects electrical data from devices installed on the PV plant and sent to the cloud, where data analysis takes place to perform ‘Root Cause Analysis’ and to predict anomalies in the system with the help of Machine Learning algorithms.
The meaning of Data Analytics has changed rapidly over the past couple of years. Now, with cutting edge technologies of Machine Learning, we can get actionable intelligence from the field data (through IoT) which is simply impossible to perform using mere statistics or regression analysis.
Using Deep Learning algorithms, we can identify field issues in solar power plants (fault classification), such as partial shading, degradation of modules, dust and dirt on modules, improper wiring, poor dynamic MPPT efficiency of inverter, etc. Statistically, it may not be possible to differentiate between these conditions.
Using temporally capable Machine Learning algorithms, predictive, pre-emptive maintenance can be performed, potentially saving permanent damage to the system. Combining these two techniques, it is possible to bring down the O&M costs drastically, as well as reduce the system downtime.
Conclusion: The industry, therefore has to rapidly moving towards adoption of smarter solutions based on IoT and advanced Machine Learning solutions.
In God We Trust, Everyone Else Bring Data to the Table – Indian Solar Industry is still at a very nascent stage of adopting this famous saying in the operations and maintenance. In today’s scenario, most of the solar plants are managed using legacy processes and systems. In case of utility solar plants, technical team is deployed and they follow certain checklists and address breakdowns as and when it happens. Cleaning labors have specific targets of cleaning ‘n’ number of tables per day. As the day ends, field engineers, plant managers churn out ‘Daily Reports’ of various activities in different formats. When Tariff rates are locked for ~2.5 to 2.6 rupees for 25 years, it puts enormous pressure on O&M costs resulting in poor maintenance leading to loss of generation. The need of the hour is ‘advanced analytics’ to introduce a ‘data driven approach’ on each and every aspect of O&M.
O&M teams & Inverter OEMs should jointly mine inverter data and come up prediction on when likely next breakdown will happen and why. This needs to be automated at inverter level or at plant SCADA/Monitoring systems. This could help creating self-healing / machine learning inverters in future. Thereby, reducing generation loss, savings on warranty costs, etc. Preventive maintenance checklists filling have to be done using Tablets/mobile phones and each check list item has to be tracked with a photo or video. Whenever a breakdown happens, the data on when failed part undergone maintenance activity should be automatically linked. This will help to understand how diligently maintenance activities are carried out.
For this, one needs to have a sophisticated software platform which interlaces pure play monitoring, O&M tasks, processes. In this way, the concept of calculating Technical manpower based on size of the project will vanish. One could manage a large 100+ MW plant with a very small team and that too on a need basis.
Like most of the emerging companies our systems have been built over a period of time by using point applications to address tactical business functionality. Mostly the required information used to be scattered across multiple data sources like SCADA, IoT collectors, ERP and internally developed applications.
We sat down with our leadership and mapped the business requirements for the next three to five years and arrived at a Technology Roadmap. We created a data platform through data mining, slicing / dicing and continuous validation through trials. Creating data lakes based on business requirements helped us create predictive analytics and business dashboards. Our achilles heel was in providing for real-time data from various sources including SCADA / PLCs in our Renewable Energy assets. This was as a critical need of the hour for business operations like energy forecasting & Scheduling which are more dynamic, time bound and needed to be accurate for actionable insights and business decisions. Usage of AI and ML in data preparation will be a definite advantage. As part of this initiative we leverage complex existing data structure for large scale process automation by involving simple servicing bots to sophisticated AI agents being deployed for data preparation and data model enhancements.
Our dashboards and energy graphs are assisting Operations & Maintenance teams to identify deviations from projected Vs actual, identifying the odd ones out and gauge the effectiveness of maintenance activities. Early predictions of machine failures will give additional advantages towards making corrective actions / replacements specifically in the Renewable Energy sector which is seasonal in nature and hence can help avoid large scale generation loss.
AI and ML are becoming the cornerstones of our future planning specifically while being used in data preparation, predictive analytics towards machine performance, for streamlining energy forecasting & scheduling operations where multiple predictions required in a day, to detect anomalies across forecasting vendors – to name a few. All these systems involve massive data collation, preparation and continual model improvements. We are on a continuous improvement journey here with a lot more possibilities to harness and improve reliability, efficiency, accuracy and speed of our service platforms.
Our initiation towards analytics readiness will give definite advantage towards reducing penalties arising due to more stringent regulatory requirements on energy forecasting. We are looking forward to plummeting generation losses with better understanding of the real-time data and fine tune our data & AI models.
Asset downtime and underperformance have always been a challenge in the renewable industry. With advanced data analytics, we can now disrupt this area with a deeper understanding of the relationship between various signals and cause of failure. This identification of failures can now be done with good lead time, thereby improving operational efficiencies and reducing costs.
In today’s world, advanced data analytics means a rapid iteration of data analysis on longitudinally and horizontally rich data. Longitudinal data represents years’ worth of data gleaned from one wind turbine, along with historical maintenance and event records. This analysis can help understand gradual performance degradation due to various reasons such as pitch/yaw system issues, higher friction during low wind conditions, and dust accumulation on blades, among others. This type of analysis is becoming a common and affordable practice, as opposed to a specialized consulting engagement, which unlocks operational performance improvement and extends the lifetime of valuable assets.
Horizontally rich data allows the operations team to view asset performance with respect to similar assets in the same geography and compare it with operating conditions. This analysis helps pinpoint acute and chronic conditions of underperforming assets. Data holds less significance if viewed in absolute terms but holds the key if users can derive insights that then can be converted into actionable items. With a diverse portfolio of renewable assets spread across the globe, it’s always helpful if companies have a top-level understanding of how their assets are performing with just one click on their computer or phone. The insights revealed also help make the decision-making process easier.
Another aspect of improving operation and maintenance is to move from scheduled maintenance to predictive maintenance. This helps customers focus their attention to areas that need immediate help, with the algorithms deciding the priority.
Scheduled maintenance can also be effectively executed with more precise power forecasting. With advancement in forecasting techniques, asset farm operators can always decide on the timing of scheduled maintenance to coincide with low power production, thereby reducing net cost.
Industries are sitting on huge amounts of data that have the power to unlock way more value than is potentially being realized from existing assets and resources. Better planning with the investment in the right technology can go a long way.
RE Analytics helps to provide insight into what has happened in the past and why particular events occurred. It helps to predict failures & Realtime performance in comparison with system design to perform predictive maintenance thereby improving the revenue & healthiness of the plant. The below are the various forms of renewable energy analytics & its advantages.
It is most commonly applied form of analytics. It is used to provide insight into what has happened in the past.
Advantage : It helps for the benchmarking
It seeks to understand why particular events occurred. This is often based on testing correlations between various relevant variables.
Advantage : It helps to find similarities / symptoms with other health components
It attempts to develop forecasts. Probability and statistics are employed to understand the likelihood of an outcome occurring based on past information. This sees the complexity of predictive analytics rising significantly relative to earlier stages of analytics.
Advantages : It helps in forecasting & performing predictive maintenance
It attempts to estimate the best responses to forecasted events. When operating within a stochastic context this requires complex computational capabilities and will be based on expected value calculations. The benefit of accurately modelling best response decisions can be significant. This form of analytics is primarily dominated by Machine Learning and Artificial Intelligence.
Advantage : it gives all the benefits of other analysis & also provides intelligence action & performance.
Reliability Centred Maintenance (RCM) leverages global best practices of Operations and Maintenance combined with advanced data analytics models using AI, ML and DL. RCM helps in decision-making related to maintenance of large assets, its components, maintenance methods and schedules in order to achieve efficient O&M operations.
In comparison with Condition Based Maintenance (CBM) which bases maintenance decisions on condition of equipment, RCM provides insights based on consequences of failure mode. RCM provides an understanding of which failure modes deserve what kind of maintenance and which task needs to be redesigned. With CBM, there can be no upper limit placed on maintenance frequency. But with RCM, it is possible to reduce or limit the number of maintenance needs to be done. However, CBM can be used effectively, with the availability of smart operational and maintenance related data analytics models.
To achieve smart operations, it is critical to track the component-wise efficiency and thereby their impact on overall asset efficiency. The overall efficiency of the asset therefore depends on the efficiency of mechanical and electrical components. And to achieve smart maintenance, it is critical to predict the failure of a component well in advance by which maintenance, replacement, cranes and spares can be planned smartly. IPPs will be able to reduce revenue loss caused due to sudden failures and add credibility to their company during investor evaluation. Through preventive maintenance or well-planned replacement activities, huge savings in maintenance costs can be achieved.
As the characteristics of each asset and each component of the asset are different, we need to have a highly scalable and intelligent model, with several layers of automation using ML & DL, to address any type of component. We can get a complete understanding of the key reasons behind the efficiency loss of an asset, anomaly detection and predict failures well in advance. This helps in taking quick real-time operational and maintenance decisions so that the asset will run at its best possible efficiency with reduced downtime.
In conclusion, the best O&M practices when combined with data analytics creates the right balance between increased revenue from generation and reduced cost of generation.