Matrix Factorization
Mathematical matrix factorization techniques decompose a rectangular matrix into three constituent matrices to analyze its structural properties and solve linear systems. In sensor calibration, singular value decomposition is applied to calculate the optimal orthogonal rotation matrix that aligns the sensor coordinate frame with the calibration platform. This method provides a numerically stable solution to the orthogonal procrustes problem.
The operation is executed during the final calculation stage of the calibration sequence.
Orthogonality Determination
Calculating the alignment matrix requires that the sensor coordinate system remains perfectly orthogonal to avoid scaling distortions. When utilizing singular value decomposition, the algorithm isolates the pure rotation from any scaling or stretching of the axes. This separation ensures that the resulting alignment matrix does not introduce artificial gain errors into the sensor calibration.
The computed rotation is represented as an orthonormal matrix.
Numerical Computation
Stability of the algorithm in the presence of measurement noise is a primary reason for its selection over other inversion methods. When executing singular value decomposition on the sensor calibration equations, the algorithm provides the singular values of the system matrix. These singular values allow for the detection of ill-conditioned systems that might arise from insufficient test orientations.
If the ratio of the largest to the smallest singular value is too high, the system is flagged as unstable, indicating that the calibration data lacks diversity. This analytical capability is used to monitor the quality of the automated calibration process.
Calibration Sourcing
Production software uses these matrix algorithms to process the calibration data of thousands of sensors per day. Since singular value decomposition is computationally demanding, it is performed on the test bench computer rather than the resource-constrained sensor microcontroller. This division of labor keeps the sensor unit cost low while ensuring high-quality calibration.