Machine Learning and Artificial Intelligence

In today’s modern complex systems, Machine Learning (ML) and more broadly Artificial Intelligence (AI) are commonly integrated into non-safety critical systems. Introducing ML technologies and techniques into safety-critical systems introduces unique challenges. These challenges are analogous to those faced in any complex system: requirements, architecture design, detailed design, integration & validation, and operation.  ML is uniquely positioned to solve hard-to-define challenges. ML has broad applications from general purpose tasks (e.g.: Conversational Bots) to domain-specific tasks (e.g.: Sensor Fusion for obstacle detection). With such broad applications, exida expects the rise of ML in safety applications across all industries. 

exida works with customers to guide how and when a ML model can be applied in a safe, reliable, and dependable manner. Our goal is to reduce risks associated with ML solutions to a tolerable level by providing feasible requirements and processes. 

exida engages in ML initiatives for (Semi-)Autonomous Vehicles, Machine, Robots, and the Process Industry. Working with partners and leaders across the globe to achieve state-of-the-art compliance strategies, engineering, advisory services, and certification with a focus on safe ML solutions.  

Applying existing and upcoming best practices, evolving ML and AI standards, along with best-in-class engineering expertise, exida is able to deliver a broad spectrum of services related to ML from data set and architecture advisory services through to certification and assessing compliance with existing and emerging standards.  

exida provides specific ML services: 

• Advisory Services 
• Analysis Services 
• Certification Services 
• Integration Services 
• Training Courses 

Request a Proposal   

Advisory Services

Advisory Services guide the ML design and development. Examples of ML Advisory Services include: 

• ML FMEA and Architecture Review: Defines failure modes, finds design problems, and identifies possible solutions with ML models early in the design process. 
• ML Process Gap Analysis: Identifies gaps between a customer’s existing ML procedures and the requirements from Functional Safety Standards and exida’s ML experts.  
• Training Datasets Specification: Provides guidance and feedback to ensure high data quality and representation in ML training datasets.  
• ML Safety Training Course: 
  • Functional Safety for Systems Incorporating AI / ML 

Analysis Services

Analysis Services analyze the model and its performance. Examples of ML Analysis Services include:

• Safety Metrics Definition: Identifies the key measures of performance and defines the level of performance that is required for the intended safety application.  
• Neural Network Architecture Analysis: Inspects and provides feedback on the safety implications of the Neural Network architecture.  
• ML Statistical Analysis: Generates a statistical estimate of the ML Models performance. This provides confidence that the ML model performs as intended based on model validation and testing.  

Certification Services

As Functional Safety Standards for ML emerge, ML Certification is an innovative application of standards-based assessments. exida is developing a strategy to certify some ML models in some applications. This exida scheme manages the risk of ML in safety critical tasks.

Integration Services

Integration Services verify the ML model in the integrated system. Integration Services include: 

• ML HARA: Identifies the hazards and risk associated with the ML model integrated into a complex system. 
• ML System Verification: Verifies the ML model and safety function in the system context. 
• ML Continuous Monitoring: Leverages automatic diagnostics to detect changes in ML performance before a failure. 

Request a Proposal