Optimizing Fleet Operations through Sensor Data Integration and Power BI Reporting

Challenges in Sensor Data Integration and Quality

Each garbage truck had hundreds of sensors collecting data on wheel speed, positions, pressures, temperatures, and switch states. However, no insights were being gained because the data wasn’t synced and lacked a consistent framework. While all the sensor data was stored in a common database, the data had issues like inconsistencies, missing values, and required changes. To make the data usable for analysis, it needed to be synced, columns had to be combined, and other transformations were necessary.

Issues with Unsynced Data

As each sensor was capturing data at different timestamps, data points across trucks or across different timeframes were difficult to compare. With such misalignment, meaningful insights and comparisons were not possible to attain.

01.

Data Synchronization and Aggregation

NeenOpal came up with a data structure to synchronize the same timestamps among the sensor data. This would allow cross-truck comparison. We stored and processed data using Snowflake through data transformation using both Python and Snowflake SQL.

02.

Correlation Analysis

We applied correlation analysis through Python to observe the interrelation between pairs of sensors. This helped the company determine how different components are performing across different trucks under different conditions.

03.

Power BI Reporting

We developed a series of interactive Power BI reports to help visualize the processed sensor data. These reports allowed users to compare sensor interactions between trucks, monitor individual sensor values and real-time data volumes, and track sensor frequencies and detect out-of-range values. They also provided insights into sensor-value distributions across trucks using kernel-density estimation and quantiles, along with timeline trends and comparative sensor values across various time aggregates such as seconds, minutes, and days.

04.

Anomaly Detection

Machine learning algorithms like isolation forest, k-means clustering have been deployed to perform anomaly detection to identify out-of-range timestamps in the data as potential early signs of fault indicators to prevent failure.

05.

Fuel Consumption Metrics

We developed custom algorithms that would calculate average fuel consumption per truck errand. This way, correlating different sensor data points, we could estimate fuel consumption during different errands, compare trucks, and benchmark against competitors.

Improved Operational Efficiency

The synchronization of data and correlation analysis enabled better decision-making, leading to a 15% increase in fleet management efficiency.

Reduced Fuel Consumption

By understanding fuel consumption trends and comparing performance across trucks, the company optimized routes and operation patterns, achieving a 10% reduction in fuel usage.

Faster Anomaly Detection & Maintenance

The machine learning-driven anomaly detection feature allowed the team to catch potential issues 75% faster, preventing costly breakdowns and minimizing downtime.