Electric Power Devices Predictive Maintenance (PdM) and Failure Prediction

Anusha C A

Anusha C A

Bengaluru, Karnataka

Engine Failure Prediction using Intel OneAPI ...learn more

Project status: Concept

oneAPI, Mobile, Networking, Internet of Things, Artificial Intelligence

Intel Technologies
oneAPI

Docs/PDFs [1]Code Samples [1]Links [5]

Overview / Usage

Power systems Predictive Maintenance (PdM) and failure prediction using Intel oneAPI can improve the performance, portability, productivity, and interoperability of PdM and failure prediction systems, by providing a single programming model for different types of hardware, enabling developers to take advantage of the full performance of different types of hardware, and simplifying the deployment of PdM and failure prediction systems across different types of power systems equipment. OneAPI enables a more efficient development process and allowing the system to handle a large volume of sensor data, making predictions in real-time, and improving the performance and reliability of power systems.

To predict device failures in power systems, machine learning algorithms are trained on historical sensor data from the devices. This data may include information on the device's usage, operating conditions, and performance metrics, such as temperature, voltage, current, Sound and vibration. The algorithms are trained using Intel oneAPI Deep Neural Network Library (oneDNN) to identify patterns and anomalies in the data that are indicative of an impending failure.

Once the model is trained, it can be used to analyze current sensor data from the device and predict when a failure is likely to occur. This prediction can be used to schedule maintenance, replace parts, or take other preventative measures to avoid the failure.

We have created a short demo video showcasing the working of the prototype -

https://www.youtube.com/watch?v=shY6TtdiLds&ab\_channel=CAAnusha

Methodology / Approach

The goal of power systems device failure prediction is to prevent equipment failures before they occur, by proactively identifying when maintenance is needed, rather than waiting for equipment to fail.

Power systems include various types of devices, such as generators, transformers, switchgears, and power transmission and distribution lines, among others. These devices are critical for the stability and reliability of the power system and their failure can cause significant disruptions to the power supply.

To predict device failures in power systems, machine learning algorithms are trained on historical sensor data from the devices. This data may include information on the device's usage, operating conditions, and performance metrics, such as temperature, voltage, current, sound, humidity and vibration. The algorithms are trained using Intel oneAPI Deep Neural Network Library (oneDNN like sklearn and tensorflow) to identify patterns and anomalies in the data that are indicative of an impending failure.

Once the model is trained, it can be used to analyze current sensor data from the device and predict when a failure is likely to occur. This prediction can be used to schedule maintenance, replace parts, or take other preventative measures to avoid the failure.

Technologies Used

The project consists of an Arduino board connected to a sound sensor, vibration sensor, DHT11 humidity and temperature sensor. The sensors are used to collect environmental data such as sound level, vibration level, temperature and humidity.

The Arduino code reads the sensor data and sends it to a computer over a serial connection. The Python code running on the computer receives the data and uses a pre-trained Intel OneAPI OneDNN(sklearn and tensorflow) machine learning model to make predictions about potential issues based on the environmental data.

The machine learning model is trained on a dataset of sensor readings taken from a variety of vehicles in different environmental conditions and engine states. The model can be trained to predict potential issues such as overheating, low oil pressure, engine misfires, or worn engine components based on the environmental data.

The system can be used in a vehicle diagnostic or maintenance application to monitor and predict potential issues with the engine or other systems based on environmental data. This can be useful for preventative maintenance or for quickly identifying and addressing issues before they become major problems.

Documents and Presentations

Repository

https://github.com/caanusha/IntelOneAPI_SmartFailurePrediction

Collaborators

1 Result

1 Result

Comments (0)