Process method for applying a machine learning algorithm to automobile software development function safety

Abstract

A process method for applying a machine learning algorithm to automobile software development function safety comprises the following seven steps: step 1, a project starting stage: adopting a machinelearning decision-making door; 2, a software security requirement stage: standardizing and describing security requirements; Step 3, a software architecture design stage: adopting a fault-tolerant design method; 4, a software development stage: collecting data; 5, a software development stage and model selection; 6, a software development stage and model implementation are carried out; And 7, a software integration stage, verification and testing. According to the method, in the development process of automobile electronic software, the property of partial standardized description is clear, analgorithm feature selection model based on various machine learning is facilitated, and then the development efficiency and safety are improved. Based on the above, the method has a good applicationprospect.

Description

technical field [0001] The invention relates to the technical fields of automotive software safety and machine learning, in particular to a process method for applying machine learning algorithms to functional safety of automotive software development. Background technique [0002] With the development of technology, there are more and more software applications for various functions of automobiles, such as advanced driver assistance systems (ADAS) and automatic driving systems (ADS). In the development process of automotive software, machine learning plays an increasingly important role, ensuring the development and application of software. Security is a major concern in automotive software development. ISO 26262 (an international safety standard for automobiles) is based on a systematic approach to safety issues from the perspective of industrial practice. This standard defines the safety life cycle of automobiles and the methods to achieve safety at each stage; The safe...

AI Technical Summary

Problems solved by technology

Because the software development of many types of advanced functions (such as automatic driving system) needs to perceive the environment, and the existing functional software cannot be fully standardized description; for example, what is the specification for identifying pedestrians? Using specifications (such as necessary and sufficient conditions) can only partially specify these conditions. In practical applications, data samples are also needed to help describe functions; since functions like perception are difficult to specify clearly, it is necessary to use machine learning-based methods to implement software Components, implementing software components by training from samples rather than procedural programming from specifications

However, in this model, functional safety assurance is hindered by the lack of specification for machine learning-based approaches.

[0005] On the other hand, all types of machine learning models contain knowledge in encoded forms that are often not very interpretable

Because, in machine learning algorithms, neural network models are usually not interpretable, more and more capabilities of machine learning models are usually at the expense of interpretability, which makes manual white-box verification methods unusable during development, such as walking inspection and inspection, creating obstacles to functional safety assurance

Images