On-Edge Aggregation Strategies over Industrial Data Produced by Autonomous Guided Vehicles

Agenda

1. Processing at the edge
2. Autonomous Guided Vehicles
3. Goal of the research
4. Analytical workflow
5. Data models
6. Aggregation methods
7. Testing environment
8. Experiments 
9. Summary

Processing at the edge

Edge computing is a computing paradigm that brings the data processing and storage closer to a place where it is needed. It allows to reduce the volume of data that needs to be send over the Internet, allows to improve reaction time to the changing state of the system and improves resilience and allows for data loss prevention where Internet connection is not reliable or not available at all most of the time.

Autonomous Guided Vehicles

Autonomous Guided Vehicle (AGV) is an autonomous robot, usually guided by markers such as wires, magnetic tapes, or uses computer vision, lasers, or GPS, for navigation. The most popular use case for such vehicles is use in industrial applications, usually for transporting heavy materials in factories.

Autonomous Guided Vehicles

Formica 1 AGV (used in our testing)

Goal of the research

The goal of the presented research is to propose and evaluate the approaches to data aggregation that could help reducing the volume of readings from AGVs, by taking advantage of the edge computing paradigm. 

Analytical workflow

The analytical workflow consists of a few steps. Firstly, the AGV client periodically retrieves the data from each AGV and persists it in a local database. A separate process is responsible for running aggregations on data retrieved from the local storage. The results from these aggregations are also persisted separately in a local database. Lastly, the third process is responsible for retrieving the aggregated data, performing optional filtering, and sending the aggregated data to the cloud for further processing.

Analytical workflow

Diagram of the aggregation workflow on the edge device

Data models

Each data point collected from AGV consists of metrics such as battery cell voltage, momentary and cumulative power, energy, current consumption, cumulative distances, and momentary frequencies. Each reading is additionally timestamped and tagged with the unique AGV identifier. The total size of a single data point from AGV is equal to 84 bytes. 

Data models

Data model of raw readings from AGVs

Data models

Data model of aggregated AGV readings

Aggregation methods

Window aggregation method

Aggregation methods

Delta updates optimization step

Testing environment

Testing environment

• Group of simulated AGVs, based on readings from real-life experiments with Formica 1 AGV, produced by AIUT LTD
• Jetson Xavier NX, serving as a database, AGV client, and an analytical engine for the AGV data
• TimescaleDB, used for persistence and analytical aggregations
• Python scripts that simulated and run all the operations

Experiments

• Tests with simulated data for 10 AGVs, based on actual sensor readings obtained from operational cycles of the Formica 1 vehicle
• For each AGV, we had 3 datasets for readings generated every 200, 500, and 1000 ms, each dataset with 1380 data points which resulted in 41400 data points for testing
• For each data set, we ran aggregations with different aggregation time windows - 2, 5, and 10 seconds
• For each scenario, we computed the total number of bytes that would have to be sent to the cloud for further processing and compared the results against the baseline of the number of bytes that would be sent if data was directly forwarded to the cloud

Results

Results

Results

Results

Summary

• Achieved data volume reduction from 2% to as much as 81.1% in tested cases
• The aggregation time window should be chosen carefully in order to not lose important data points (e.g. spikes)
• Using the delta updates method can be very effective in data volume reduction if some of the values are changing slowly/rarely in comparison to data acquisition frequency
• There is an additional benefit to buffering data on the edge device of avoiding data loss in case of network failures