Project Name
Detection of Short Cycling (Faulty Refrigerators) of Compressor With Machine Learning
Overview
Our client was a leading provider of solutions for refrigeration systems. They were facing challenges that included compressor short cycling in refrigerators which led to an increase in power consumption and potential damage to the compressor. The client was looking for a solution that would help them to detect and prevent these short cycles.
(A short cycle usually happens when the cooling becomes shorter than the usual time, and because of this, the compressor will turned on and off more as per the situation)
Challenges
Our client were facing multiple challenges that include:
- Dealing with massive amounts of unstructured data needs a robust data extraction and cleaning strategy.
- Facing issues in creating rules for average and short cycles domain expertise and understanding of compressor behavior.
- While deploying CodeBuild, it encounters problems with integrations and dependencies.
- The major issue is setting up ECS, fargate clusters, and task definitions with network and security.
Our Solution
We have provided them with a robust solution that includes:
- We helped them to do data extraction that extracted AMPS values from unstructured data and organized them based on date.
- Our team imported the model XGBRegressor() class from the XGBoost library with the hyper-parameters passed as arguments.
- They had done data cleaning that removed noisy data through filters, ensured data quality, and handled missing values by preparing the dataset.
- We had set up 2 different pipelines using AWS CodePipeline and CodeBuild. Even, our team had created Docker images and stored them in AWS ECR for both model training and prediction.
- Successfully, our team implemented a workflow to calculate the number of short cycles with an algorithm on segment counts.
Short Cycle:
Normal Cycle:
Data Flow Diagram
Conclusion
At last, we conclude that our team had effectively addressed all the client challenges by utilizing Advanced machine learning techniques that ensured the accurate detection of compressor short cycling. Moreover, the integration of our ML model for the client’s infrastructure is coupled with automated pipelines. Hence, our client has a robust and scalable solution for ongoing monitoring and prevention of compressor short cycles in refrigeration systems.
