About Our Product and Best Practices

Conducting a high-quality FMEA has historically presented a difficult trade-off between depth and speed. Engineers often face the choice of spending weeks meticulously building failure chains or rushing the process to meet deadlines, potentially leaving important risks undiscovered.

Our application is designed to resolve this tension. By leveraging advanced generative AI within a specialized engineering framework, we automate the heavy lifting of structure creation, failure mode identification, and risk mapping, delivering a comprehensive analysis in a single run.

An Agentic Workflow: The Virtual Engineering Team

Unlike standard AI tools that simply "predict text," our application utilizes an agentic workflow. This means the AI simulates a cross-functional team of experts working in collaboration, rather than a single entity attempting to perform all tasks at once.

Instead of generating a flat list of risks, the system employs a multi-stage process with built-in quality gates. It deconstructs your documents to understand system architecture, generates logical failure chains, and rigorously self-corrects against engineering standards before the output is presented.

This approach enables the system to generate up to 200 technically distinct, logically consistent failure modes in a single run, with a depth that mirrors a facilitated engineering workshop.

Best Practices: Optimizing Your Output

The quality of the generated FMEA is directly proportional to the quality, clarity, and relevance of the documents you provide. The AI does not invent technical knowledge; it grounds its analysis strictly in your inputs.

File Formats

The application supports PDF, CSV, TXT, XLSX, and image files. However, PDF format is strongly recommended.

Modern multimodal AI models do not only read text, but they also interpret visual structure. PDFs allow the system to extract information from:

  • Block and boundary diagrams
  • Schematics and drawings
  • Tables, callouts, and annotations
  • Page structure and hierarchy

This visual understanding significantly improves system decomposition, interface identification, and functional accuracy.

Capacity Limit

To preserve deep contextual reasoning, each analysis run currently supports up to 100,000 tokens, equivalent to approximately 300-400 pages of typical technical documentation.

Uploading fewer, higher-quality, and well-structured documents usually produces better results than uploading many loosely related files.

For Design FMEA (DFMEA)

Focus on documents that define system behaviour, boundaries, and physics:

  • Requirements specifications
  • Architecture and block diagrams
  • Design descriptions and schematics
  • Interface definitions
  • Bill of Materials (BOM)
  • Validation and verification plans

These documents allow the AI to ground functions, interfaces, prevention controls, and detection logic in objective design intent.

For Process FMEA (PFMEA)

PFMEA requires documentation that connects manufacturing execution to product characteristics:

  • Process flow diagrams
  • Work instructions and SOPs
  • Control plans
  • Engineering drawings and tolerances
  • Design FMEA (for severity linkage)
  • Scrap history, deviations, and lessons learned

This enables accurate 4M analysis and realistic prevention and detection control identification.

Output

Standards Alignment

All outputs are delivered in an AIAG & VDA aligned Excel structure using the harmonized 7-Step approach and Action Priority logic.

While originally developed for automotive applications, this methodology has become the de facto standard for technical risk analysis across aerospace, medical, industrial, and high-reliability industries.

Using this structure ensures that your FMEA remains review-ready, auditable, and logically consistent.

How to Interpret the Results

The generated output is a complete, structured FMEA draft.

Each row represents a full failure chain, connecting:

  • Structure (system or process hierarchy)
  • Functions and requirements
  • Failure Causes → Failure Modes → Failure Effects
  • Current controls
  • Risk ratings and Action Priority
  • Recommended optimization actions

The application generates approximately the requested number of technically distinct failure modes, distributed across the system or process structure to ensure broad and balanced coverage.

The output should be treated as a high-quality engineering draft, intended to accelerate structured FMEA development rather than replace professional judgment.

Evidence-Based Controls

The AI searches your uploaded documents to locate objective support for:

  • Current Prevention Controls
  • Current Detection Controls

The Remarks column explicitly shows the source and reasoning used. This allows rapid validation and easy identification of assumptions or gaps.

Optimization Actions

Preventive and detection actions are generated as engineering suggestions based on the failure physics. These should be reviewed and adjusted by the engineering team to confirm:

  • Technical feasibility
  • Economic justification
  • Organizational applicability

Administrative Fields

Project management fields (responsibility, dates, status) are intentionally left blank. These must be assigned through your internal processes.

The Human in the Loop

This application applies advanced self-review logic and quality gates to produce a disciplined FMEA draft. However, no automated system can fully replace engineering judgment.

The output should always be:

  • Reviewed by qualified subject matter experts
  • Adjusted to reflect final design and process intent
  • Approved through your standard FMEA governance process

Practical Tips for Efficient Use

Simplicity of Use

The application is intentionally designed to remain simple and focused. A typical workflow consists of only four steps:

  • Select the FMEA type (DFMEA or PFMEA).
  • Upload the relevant technical documents.
  • Define the desired number of failure modes.
  • Click Run FMEA Analysis.

No additional configuration is required.

Selecting the Number of Failure Modes

The requested number of failure modes is the primary cost and computation driver.

The value is adjustable up to 200 failure modes and should be selected based on:

  • System or process complexity
  • Level of decomposition required
  • Project phase and risk exposure

For early concept phases, fewer failure modes may be sufficient. For detailed design or manufacturing release, higher values are usually justified.

Document Format and Token Usage

Whenever possible, PDF format should be used. PDF files preserve layout, diagrams, tables, and hierarchy, which significantly improves system understanding.

The application supports up to 100,000 tokens per analysis run, which is sufficient for most real-world projects. Token usage per document is visible to the user.

If the token limit is approached, users can:

  • Remove less relevant documents
  • Shorten repetitive or administrative content
  • Replace low-value files with higher-quality technical sources

This allows users to optimize input quality without sacrificing essential information.

Analysis Duration

FMEA generation time depends on:

  • Total document volume
  • Document complexity
  • Requested number of failure modes

In large analyses, generation can take up to one hour. This reflects the multi-stage reasoning and internal quality review process applied during analysis.

Credit Usage Policy

Credits are charged only when the delivered Excel file contains more than 75% of the requested failure modes.

If the system generates more complete FMEA entries than requested, no additional credits are charged.

This ensures that users only pay for usable, complete engineering output.

Importance of Document Quality

Document quality is the single most important factor influencing FMEA quality.

Clear, structured, technically meaningful documents lead directly to:

  • Better failure chain logic
  • More realistic controls
  • More accurate ratings
  • More practical optimization actions

Poor, incomplete, or inconsistent documentation will limit output quality regardless of AI capability.

Support and Feedback

If you have questions, require guidance, or wish to share feedback, your input is always welcome and our team is available to support you.