Powertrain Causal Analytics for Real-Time Sensor Calibration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing powertrain management systems face challenges in achieving real-time, granular, and fine-tuned monitoring and management due to limitations in sensor calibration, reliance on biased 'big data' models, and the need for extensive offline testing, which hinders adaptability to dynamic conditions and environmental variations.
Innovation Solution
Implementing causal analytics through experimental signal injections and collaborative learning across vehicles to optimize powertrain control parameters in real-time, leveraging a small data size and low computing power, allowing vehicles to share knowledge and adapt to local conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If extensive offline testing and big data models are used for powertrain calibration, then model robustness under broad conditions is improved, but development time and cost increase significantly
Solution Approach 1:
The system performs preliminary calibration actions during manufacturing by injecting test signals and measuring sensor responses before the vehicle is deployed. This preliminary characterization of sensor behavior under known conditions eliminates the need for extensive post-deployment testing and big data collection, directly reducing development time while ensuring model robustness from the start
Solution Approach 2:
The calibration system is self-sufficient by using built-in signal injection capabilities and onboard sensors to automatically characterize sensor behavior without requiring external testing facilities or large datasets. Each vehicle calibrates its own sensor system independently during a brief test period, eliminating dependency on extensive offline testing infrastructure
2Measurement precision
If sensor calibration is performed through extensive bench testing, then measurement accuracy is improved, but calibration complexity and cost increase
Solution Approach 1:
Each sensor system performs its own calibration by injecting known test signals and measuring its own response characteristics. This self-calibration approach eliminates the need for complex external calibration equipment and procedures, achieving high measurement accuracy while simplifying the calibration process to something that can be done automatically during vehicle operation
Solution Approach 2:
The calibration process extracts only the essential characterization data needed for accurate sensing by injecting specific test signals and measuring responses under controlled conditions. This extracted minimal dataset is sufficient for calibration, eliminating the need for extensive bench testing across all possible operating conditions and thereby reducing calibration complexity
3Measurement precision
If control loops are used for feedback correction, then sensor accuracy compensation is improved, but functionality is limited to single-control mappings
Solution Approach 1:
The system merges data from multiple sensors and multiple control parameters into a unified causal model that captures complex relationships between controls and outcomes. This integrated approach allows simultaneous optimization of multiple controls (e.g., spark timing, injection timing, valve timing) based on their combined effect on performance metrics, rather than treating each control independently as in traditional feedback loops
4Measurement precision
If machine learning models assume system stationarity, then historical data accuracy is improved, but adaptability to dynamic conditions deteriorates
Solution Approach 1:
The causal model is designed to be dynamic rather than static, continuously updating its understanding of sensor-behavior relationships as new data becomes available. The model adapts to changing operating conditions by learning new causal relationships online, allowing it to maintain accuracy under dynamic conditions while still leveraging historical data patterns when appropriate
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Methods for management of a powertrain system in a vehicle. The methods receive data or signals from multiple sensors associated with the vehicle. Optimum thresholds for classifications of the sensor data can be changed based injecting signals into the powertrain system and receiving responsive signals. Expected priorities for the sensor signals can be altered based upon attributes of the signals and confirming actual priorities for the signals. Look-up tables for engine management can be modified based upon injecting signals into the powertrain system and measuring a utility of the responsive signals. The methods can thus dynamically alter and modify data for powertrain management, such as look-up tables, during vehicle operation under a wide range of conditions.