Home>Case Studies

Case Studies

Case Studies

Pattern Recognition in Diagnostic Trouble Code
THE PROBLEM

A large American Automobile Manufacturer wanted to use Diagnostic Trouble Codes for anomaly detection and indicator of potential part failure in future for its four-wheeler business to do preventive maintenance to save costs

THE SOLUTION

10,000 DTC series codes were studies for pattern analysis to validate existing sequence of DTCs before failure. Parallel processing was done on SPARK SCALA, .9 mn vehicle data and 1.5 Petabyte of data analyzed by applying random forest algorithms

THE BENEFITS
  • 96% Accuracy
  • Improved customer experience
  • Decrease the warranty costs
  • Significant dollar savings
Yaw Optimization – Wind Turbines
THE PROBLEM

Independent Power Produces (IPP;’s) currently spend on costly LiDAR technology to measure the degrees of YAW MISALIGNMENT for WIND Turbines. High cost, delayed results and lack of scalability are some of the challenges faced.

THE SOLUTION

We developed a digital lidar based on real time SCADA data applying machine learning techniques. Multiple models were developed to pick Random Forest as the best model. The client can now see real time yaw misalignment based on last 1 months.

THE BENEFITS
  • The impact to the bottom line? An increase of $5,700/MW/Yr
  • Power curve improvement by 5-10%
Defect Detection in Oil & Gas Pipeline
THE PROBLEM

Identifying defects in Oil & Gas pipelines running into hundreds of kilometers is a challenge for companies. The data captures through pipeline inspection gauge is huge and identifying dents, ovality and corrosion if of prime interest for oil and gas players.

THE SOLUTION
Machine learning model were developed to help client correctly identify the anomaly. Reducing the false positives by 89%. All the data was processed in Azure cloud and the models were developed using Python.
THE BENEFITS
  • Identified 100% true positives
  • Achieved 89% reduction in false alarms; significantly reduced pipeline inspection costs; reduced analyst time by 70%
Energy Forecasting
THE PROBLEM

Providing schedule of energy forecast is a mandatory compliance for IIP’s and they often face heavy penalty for going beyond the forecast. The client was facing a high penalty when they approached us with the problem.

THE SOLUTION
We developed models based on xgbost algorithm to help the client accurately forecast the energy production based on downtime information, weather data and data from SCADA systems
THE BENEFITS
  • 50% reduction on penalty cost for the client.
Predictive Asset Maintenance - WIND
THE PROBLEM

Due to a competitive market in the renewable business its essential for the IPP’s to have self O&M capacity. The client has outsources O&M to OEM’s and it was impacting their bottom line.

THE SOLUTION
We developed anomaly detection model for the client based on Gaussian Algorithms. The visualization was developed on Grafana. The model shows anomaly at component and sub-component level for wind sites
THE BENEFITS
  • $500k monthly O&M reduction
  • Gain in power curve by 2%
Predictive Asset Maintenance - SOLAR
THE PROBLEM

Clients existing operating platform has solar monitoring and alarm alerting capacity. However unable to know the equipment fault and anomaly insights for its assets.

  • Know the production chain loss for its assets
  • Know the production performance insights of its assets
THE SOLUTION
An advanced dashboard to provide real time information & empower with statistical, probabilistic and data analytics capability to tackle issue in real time. Solution has capability to
  • Estimate energy loss analysis for energy production chain; Soiling Loss etc..
  • Production Performance Analysis
THE BENEFITS
  • Reduce O&M manpower cost
  • Reduce Unplanned Asset Downtime
  • Understand sites solar production chains losses and recommend corrective action
  • Overall O&M Cost was reduced from $9.5k /MW to 5.4
  • Improved energy yield by 0.3% to 1% through analytical insights
Open chat