Deep Neural Network IMU Sensor Compensation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing IMU sensor compensation techniques require user understanding of sensor physics and environmental effects, leading to inaccuracies and reliability issues.
Innovation Solution
A deep neural network-based methodology for IMU sensor compensation that learns to transform raw sensor data into enhanced compensated signals without requiring user knowledge of sensor physics or environmental conditions, using a stacked neural network architecture and training with ground truth data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If traditional least square or polynomial fitting techniques are used for IMU sensor compensation, then the compensation process is simpler to implement, but the accuracy and reliability of sensor compensation deteriorates
Solution Approach 1:
The patent replaces traditional mathematical compensation methods (least square fitting, polynomial fitting) with a deep neural network-based electronic/software system. The DNN learns complex nonlinear relationships between sensor outputs and environmental conditions through training, substituting the mechanical/mathematical fitting process with an intelligent algorithm that achieves superior compensation accuracy without requiring explicit physical models.
Solution Approach 2:
The patent transforms the compensation approach by changing from fixed parametric models (scale factor, bias, misalignment matrices) to adaptive parameters learned by the DNN. The neural network dynamically adjusts compensation parameters based on environmental conditions, allowing the system to adapt to varying temperatures, pressures, and other environmental factors that affect sensor performance.
2Ease of operation
If traditional TCAL method with least-squares regression is used, then the compensation process is more straightforward, but user understanding of sensor physics and environmental effects is required
Solution Approach 1:
The deep neural network performs self-learning and self-compensation by automatically processing sensor data and environmental condition inputs. The DNN training process autonomously discovers the relationships between environmental factors and sensor behavior without requiring user intervention or understanding of the underlying physics. The system serves itself by continuously learning and adapting to environmental variations.
Solution Approach 2:
The patent introduces the deep neural network as an intermediary layer between the raw sensor data and the compensated output. This intermediary automatically handles the complex transformations and compensations that would otherwise require user understanding of sensor physics, shielding the user from complexity while delivering accurate compensated results.
3Measurement precision
If deep neural network-based compensation is implemented, then sensor compensation accuracy and reliability improves significantly, but the computational complexity and training requirements increase
Solution Approach 1:
The patent performs the computationally intensive deep neural network training in advance during a calibration phase using controlled environmental conditions and known reference data. Once trained, the DNN model is deployed for real-time compensation with minimal computational overhead. This preliminary action separates the heavy computational work from the operational phase, enabling high accuracy during actual sensor usage without burdening the real-time system.
Data Source
AI summary
An Inertial Measurement Unit (IMU), method, and navigation system are disclosed. For example, the method includes receiving a plurality of sensor values from an IMU, loading the plurality of sensor values and a plurality of true sensor values into a deep learning algorithm, and training the deep learning algorithm to enhance the accuracy of the plurality of sensor values utilizing the plurality of true sensor values.