How Adani Renewables adopted an AWS microservices architecture for transformer health analytics
Adani Group is a diversified organization in India with combined revenue of $15 Bn comprising 6 publicly traded companies with a transport logistics, energy utility and infrastructure portfolio that has a pan-India presence. Adani owes its success and leadership position to its core philosophy of ‘Nation Building’ driven by ‘Growth with Goodness,” a guiding principle for sustainable growth. Adani is committed to increase its ESG (Environmental, Social and Governance) footprint by realigning its businesses with emphasis on climate protection and increasing community outreach through its CSR (Corporate social responsibility) program based on the principles of sustainability, diversity, and shared values.
Scope of transformer health needs at Adani
Adani has business interests across the entire electricity value chain. From manufacturing of solar panels, to generation (thermal, solar & wind), to transmission and distribution. Transformers play a critical role across the Adani value chain:
Generation: The transformers used here are primarily power transformers. They are typically 5 MVA or above with voltages 33KV and above. In renewable, inverter duty transformers, a special type of transformers, are also used.
Transmission: The transformers used here are power transformers. They are typically 5 MVA and above with voltages 132KV, 220KV, or above. Transmission also uses reactor transformers to improve system efficiency.
Distribution: The electricity from transmission lines is transferred to consumers through a distribution network. The transformers used here are distribution transformers. They are small transformers typically in the range of 10KVA – 2500KVA.
Transformers are well maintained, static devices with few moving parts. Consequently, they have low failure rates and long life. However, since they play critical role in the electricity value chain, any failure can cause a severe disruption. Transformers used in generation and transmission are typically connected to SCADA (Supervisory Control and Data Acquisition), but the monitoring is limited to alarms and trips. In addition, data from periodic inspections is maintained in spreadsheets. Over the years, multiple rule-based and engineering first principle-based models have been developed to categorize the health of transformers and to determine residual life. However, in the past, Adani has made limited or no use of historical data to analyze patterns, determine health of a transformer, flag anomalies, and predict transformer failures.
Predicting transformer health & failures is a complex task involving capture of real-time data of 40+ data points from each transformer, data quality validation, data analysis, and benchmarking based on vendor manufacturer. Using such an approach companies can diagnose or predict health trends, faults, and predictive maintenance needs. Given the large portfolio of transformers across Adani Energy businesses, the availability of years of domain knowledge and data, and because of the high cost of transformer failures, this is a problem worth solving.
Adani’s transformer health analytics solution
Adani Energy to build a “Transformers Health App” solution with data integration, health monitoring, diagnosis, fault prediction and maintenance prescription feature modules, loosely coupled to the other and built as a separate microservices so they can be offered independently or collectively.
To start, Adani Energy built a transformer health analytics minimum viable product (MVP) for inverter duty transformers used in renewable energy. This MVP included data integration, health diagnosis and monitoring features with platform scale features as mentioned below.
- Able to analyze & present both near-real time and batch data / insights.
- Multi-tenant (customer/ transformer).
- Each transformer model has its unique digital twin with a dedicated ML/AI engine.
- Historical data for each transformer retained.
The transformer health analytics MVP with microservices architecture was built in 3 Weeks with a 4 member team who collaborated through 22+ virtual meetings, each having duration of 1 – 2.5 hours. The team included a program manager, domain experts, lead engineer and data scientist from Adani Group, and a solutions architect from AWS.
The transformer health analytics MVP solution visualization for one of the key transformer data point, Winding Temperature, is as shown:
A Python script with logic to fetch and validate 40+ data points for each transformer runs on AWS Batch managed instances at a scheduled intervals. AWS Batch makes it easy and efficient to run hundreds of thousands of batch computing jobs on AWS. The Python script stores the batch processed data in company-specific S3 Bucket, or Amazon Simple Storage Service (Amazon S3), which provides durable and scalable object storage for staging raw data. Once the file is created, Amazon S3 invokes AWS Lambda function asynchronously with an event that contains details about created file. AWS Lambda lets you run code at scale without provisioning or managing servers. This input scheduler Lambda function retrieves the input data from Amazon S3 and invokes a REST API in Amazon API Gateway, which is backed by another Lambda function to invoke Amazon SageMaker inference endpoint. Amazon SageMaker is a fully managed service that provides ability to build, train, and deploy machine learning (ML) models quickly. This transformer anomaly detection machine learning model and process returns a result and stores it in Amazon Aurora Serverless DB (MySQL compliant). SQL database is used to store multi-tenant master data for customers and transformers, respective model inference results and historical data for further analytics and visualization. Using Amazon Aurora Serverless, you can run database in the cloud without managing any database instances, a simple and cost-effective option for infrequent, intermittent, or unpredictable workloads. From there, Amazon QuickSight is used to visualize Transformer anomalies results by respective users of each company. Amazon QuickSight is a cloud-powered business intelligence service that lets you easily create and publish interactive dashboards that include ML Insights. The transformer anomaly detection machine learning model to detect anomalies from inverter duty transformers data was built and trained in-house at Adani Group. Data scientist used Amazon SageMaker Notebook instances to deploy and create endpoints for custom machine learning models on AWS. AWS Developer Tools were used by the Lead Engineer and Data Scientist to develop and automate deployment of Python scripts through the DevOps pipeline.
The “Transformers Health Analytics” MVP Solution implementation on AWS helped Adani Group understand their end-to-end microservices architecture development and deployment with a multi-tenant scenario. Adani found building microservices with the help of AWS has the potential to reduce the development to deployment cycle from months to weeks.
Adani Energy is planning to add fault prediction and maintenance prescription feature modules as microservices in the “Transformers Health App” Solution. They will use Amazon SageMaker services for labelling, building, and training machine learning models for the remaining types of transformers. They will also scale out the solution to include various data ingestion scenarios and different transformer models. Microservices architectures make applications easier to scale and faster to develop, enabling innovation and accelerating time-to-market for new features. Learn more about microservices from What are Microservices and Microservices on AWS whitepapers. Visit AWS Power & Utilities for more information.