Discover how AI is revolutionising manufacturing quality control in 2026. This guide, co-authored with Lean Six Sigma expert Srinivas T V, shows you how to generate accurate Process FMEAs in minutes u
Excellent guide! I followed Option 2 to create a PFMEA for a new activity we are planning to implement, and the step-by-step approach made the process fast and straightforward. It highlighted key points with minimal effort required for fine-tuning. I found this extremely practical and useful, and I’m looking forward to sharing it with my circle—it’s truly awesome!
Brilliant walkthrough of the FMEA automation process. The Plus/Minus/Zero framwork for failure mode identification is probaly the most underrated part here becuase it forces structured thinking even when using generative models. I dunno if the 90% accuracy claim holds across all machining operations, but the template-based appraoch definitely reduces that initial blank-page friction most quality teams face.
Went through the blog - it will be definitely helpful in driving AI adoption in manufacturing. Thank you for sharing. FMEA can be an exhausting activity and any automation gives a good jump start -it will be aleays better to start with ~90% accurate FMEA, than to start with a blank template.
I have created ~12 PFEMAs in last 12 months. In recent months, I adopted LLMs (ChatGPT) for generating PFMEAs.
Thought of sharing two observations - please correct me if I am missing something.
1. LLMs typically give 1-1 mapping of FailureMode-Effect-RootCause-Control. We can edit the prompt or the GPT to create a 1-to-many mapping; e.g. 1 failure mode can have multiple effects and so on.
2. As per my understanding, GPTs (in OpenAI) are not equipped to retain context. So they would not learn over time when exposed to many FMEAs - hence giving robust prompts on ChatGPT is a better choice.
Thanks, Dhanwantri, for sharing your thoughts. Great to hear that you’re already using AI for FMEA and achieving nearly 90% accuracy — that’s impressive.
Let me address your questions:
1. Regarding Failure Modes and Effects
A single failure mode should ideally have one unified failure effect. If multiple effects exist, it’s best to combine them into a single consolidated effect.
However, the failure causes can (and often do) differ. These should appear in separate rows; this is a bit tricky to do. Please try Option B above. What I remember is that we have structured it to ensure each failure cause is presented in its own row. Do let me know if this approach worked for you.
2. On Retaining Context in OpenAI / GPT Models
GPT models retain context within the same chat. The simplest way to manage multiple FMEAs is to:
Create separate chats for each FMEA, or
Clearly name and label each FMEA within the chat.
You can then recall or continue any FMEA whenever needed. This works reliably in the paid versions, though I’m not fully sure about the limitations in the free version.
Happy to support further—feel free to share your next observations or questions!
Excellent guide! I followed Option 2 to create a PFMEA for a new activity we are planning to implement, and the step-by-step approach made the process fast and straightforward. It highlighted key points with minimal effort required for fine-tuning. I found this extremely practical and useful, and I’m looking forward to sharing it with my circle—it’s truly awesome!
Brilliant walkthrough of the FMEA automation process. The Plus/Minus/Zero framwork for failure mode identification is probaly the most underrated part here becuase it forces structured thinking even when using generative models. I dunno if the 90% accuracy claim holds across all machining operations, but the template-based appraoch definitely reduces that initial blank-page friction most quality teams face.
Thank you
Excellent thought 🤔,
Dear Srini
Went through the blog - it will be definitely helpful in driving AI adoption in manufacturing. Thank you for sharing. FMEA can be an exhausting activity and any automation gives a good jump start -it will be aleays better to start with ~90% accurate FMEA, than to start with a blank template.
I have created ~12 PFEMAs in last 12 months. In recent months, I adopted LLMs (ChatGPT) for generating PFMEAs.
Thought of sharing two observations - please correct me if I am missing something.
1. LLMs typically give 1-1 mapping of FailureMode-Effect-RootCause-Control. We can edit the prompt or the GPT to create a 1-to-many mapping; e.g. 1 failure mode can have multiple effects and so on.
2. As per my understanding, GPTs (in OpenAI) are not equipped to retain context. So they would not learn over time when exposed to many FMEAs - hence giving robust prompts on ChatGPT is a better choice.
Thanks, Dhanwantri, for sharing your thoughts. Great to hear that you’re already using AI for FMEA and achieving nearly 90% accuracy — that’s impressive.
Let me address your questions:
1. Regarding Failure Modes and Effects
A single failure mode should ideally have one unified failure effect. If multiple effects exist, it’s best to combine them into a single consolidated effect.
However, the failure causes can (and often do) differ. These should appear in separate rows; this is a bit tricky to do. Please try Option B above. What I remember is that we have structured it to ensure each failure cause is presented in its own row. Do let me know if this approach worked for you.
2. On Retaining Context in OpenAI / GPT Models
GPT models retain context within the same chat. The simplest way to manage multiple FMEAs is to:
Create separate chats for each FMEA, or
Clearly name and label each FMEA within the chat.
You can then recall or continue any FMEA whenever needed. This works reliably in the paid versions, though I’m not fully sure about the limitations in the free version.
Happy to support further—feel free to share your next observations or questions!