MENU

Mobile app for anomaly detection

Anomaly detection and outage prediction have critical importance in power generation services. Our Fortune 500 client needed a solution for engineers to find malfunctioning machines and other assets. The goal of the project was to create a mobile application that was able to identify malfunctioning assets based on thermal imagery. In this scenario, heat loss was the main symptom of these anomalies.

2013-05-30

Anomaly detection and outage prediction have critical importance in power generation services. Our Fortune 500 client needed a solution for engineers to find malfunctioning machines and other assets. The goal of the project was to create a mobile application that was able to identify malfunctioning assets based on thermal imagery. In this scenario, heat loss was the main symptom of these anomalies.

Our team consisted of both software engineers and data scientists. We tackled the problem using the following technology stack:

  • Amazon EC2
  • Java
  • Python
  • Node.js
  • Ionic
  • Angular
  • FLIR SDK

We followed a lean approach to adapt to the client’s evolving business requirements.  We shipped several MVPs and developed new features and updates as the requirements became increasingly elaborate. To keep the product cross-platform, we developed a hybrid mobile application using the Ionic framework. This enabled us to ship an app that works on both iOS and Android devices.

Applied Data Science Approaches

  • Supervised and unsupervised deep learning techniques using
    • Discriminant analysis
    • Support vector machine
    • Color histograms
  • Model parameter optimization
  • Model ensembling
  • Accuracy evaluation with different measures

Challenges tackled

  • Image pre-processing
  • Cluttered image backgrounds removed to focus only on the object of interest
  • Raw temperature metadata extracted from the images’ binaries to get absolute values
  • Images adjusted to be indifferent for various viewpoints
  • Computational time reduction by parallelization
  • Handling accuracy – computational time tradeoff

The application assists field service engineers in finding heat losses with greater accuracy. They can take thermal images and upload them for analysis. When the algorithm finds an anomaly, the service sends a notification identifying the asset. The engineer can take action and flag the image with the corresponding tag. Other engineers working in the plant can browse these images as well. All the data science techniques and UX solutions serve the purpose of finding anomalies with more ease. The assets are searchable across several asset hierarchy levels and across several sites. We thus provided the client a tool that helps engineers collaborate with each other and service their power plants.

Comments are closed.