Abstract:Ensuring fault tolerance of Highly Automated Vehicles (HAVs) is crucial for their safety due to the presence of potentially severe faults. Hence, Fault Injection (FI) testing is conducted by practitioners to evaluate the safety level of HAVs. To fully cover test cases, various driving scenarios and fault settings should be considered. However, due to numerous combinations of test scenarios and fault settings, the testing space can be complex and high-dimensional. In addition, evaluating performance in all newly added scenarios is resource-consuming. The rarity of critical faults that can cause security problems further strengthens the challenge. To address these challenges, we propose to accelerate FI testing under the low-rank Smoothness Regularized Matrix Factorization (SRMF) framework. We first organize the sparse evaluated data into a structured matrix based on its safety values. Then the untested values are estimated by the correlation captured by the matrix structure. To address high dimensionality, a low-rank constraint is imposed on the testing space. To exploit the relationships between existing scenarios and new scenarios and capture the local regularity of critical faults, three types of smoothness regularization are further designed as a complement. We conduct experiments on car following and cut in scenarios. The results indicate that SRMF has the lowest prediction error in various scenarios and is capable of predicting rare critical faults compared to other machine learning models. In addition, SRMF can achieve 1171 acceleration rate, 99.3% precision and 91.1% F1 score in identifying critical faults. To the best of our knowledge, this is the first work to introduce low-rank models to FI testing of HAVs.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is: **How to efficiently conduct Fault Injection (FI) tests on Highly Automated Vehicles (HAVs) with limited test resources to ensure their fault tolerance and safety**.
Specifically, the paper proposes solutions to the following challenges:
1. **High - dimensional test space problem**: Due to the very complex combinations of driving scenarios and fault settings, the dimension of the test space is extremely high. This makes it extremely difficult to comprehensively cover all possible test cases.
2. **Sparse data problem**: Only a limited number of simulation test results can be obtained in actual tests, and these results form a sparse data set. Therefore, a method is needed to predict untested situations, especially those critical faults that may lead to safety problems.
3. **New scenario prediction problem**: In order to reduce resource consumption, it is necessary to directly predict the performance in new scenarios based on the knowledge of historical scenarios. However, existing surrogate models are prone to over - fitting the training set, resulting in poor performance in new scenarios.
4. **Rare critical fault problem**: Critical faults (i.e., faults that may lead to safety violations) are very rare in the overall space and usually appear as some local patterns in the test space. If the surrogate model does not consider these local rules, some critical faults may be missed.
To solve these problems, the paper proposes a method based on Smoothness Regularized Matrix Factorization (SRMF). By introducing low - rank constraints and smoothness regularization, SRMF can effectively capture the correlations in the high - dimensional test space and better predict new untested scenarios and locate critical faults. Experimental results show that SRMF is superior to other machine - learning models in terms of prediction accuracy and efficiency.
### Main contributions
- **For the first time, apply the low - rank matrix decomposition model to the fault injection test of highly automated vehicles**, considering the high - dimensional correlations of the test space.
- **Design three types of local smoothness regularization** and integrate them into the matrix decomposition framework to better predict new scenarios and locate critical faults.
### Formula summary
- Definition of the fault matrix \(F\):
\[
F=\begin{bmatrix}
f_{00}&f_{01}&\cdots&f_{0J}\\
f_{10}&f_{11}&\cdots&f_{1J}\\
\vdots&\vdots&\ddots&\vdots\\
f_{I0}&f_{I1}&\cdots&f_{IJ}
\end{bmatrix}
\]
- Definition of critical faults:
\[
F_{\text{critical}} = \{ f_{ij} \mid x_{ij} > 0 \}
\]
- Sampling operator \(P_\Omega(X)\) of observations:
\[
[P_\Omega(X)]_{i,m}=\begin{cases}
y_{i,m},&\text{if }(i,m)\in\Omega,\\
0,&\text{otherwise}
\end{cases}
\]
- Objective function of low - rank matrix decomposition:
\[
\min_{W,H}\frac{1}{2}\|P_\Omega(X - W^\top H)\|_F^2+\frac{\rho}{2}(\|W\|_F^2+\|H\|_F^2)
\]
- Optimization objective with added smoothness regularization:
\[
\min_{W,H,T_u}\frac{1}{2}\|P_\Omega(X - W^\top H)\|_F^2+\frac{\rho}{2}(\|W\|_F^2+\|H\|_F^2)+\frac{\lambda_1}{2}\|H\Ps