It depends on what sensors are built in. A simple kind of input is whether the operation of smoke extraction is automatically triggered or manually, or whether the fire squad was called. A temperature sensor should be there and smoke density should be measured with light. Optional sensors could be for CO, CO2, ammonia, chlorine. For the use of AI, actors must be there to do something useful, such as increasing the pump power or raising alarms, or get into dialog with a human.

Regards,

Joachim

Joachim Pimiskern Thank you for your response. In our system, we have 2 zones (right and left). The fan motors turn to the right if smoke is detected on the right side, otherwise, they turn on the left. They can not operate the 2 directions simultaneously.

I am wondering if we could integrate a camera with AI detection to identify and count the number of people in each zone. The AI then could decide which side to activate based on the number of people present in the area, optimizing smoke and ensuring safety.

What do you think?

Cameras are really useful. Currently, you can use detection models such as Yolo (choose your variant) which can be easily finetuned (if can't find one that has already been finetuned for the task. You can not only detect smoke but also where it's coming from, possibly detect and count people in the space, etc. A human in the loop could be useful: notify somebody that can verify the situation and ensure no false alarms and wrong detections. Reporting and/or thresholding model confidence might be important to avoid problems with the system's capabilities. Kaggle, Huggingface and Paperswithcode might be useful to find smoke detection models

It's a moral dilemma for an AI. Either save the people in one room or the other. I find the type of architecture strange. Why don't you lead the smoke out of the rooms?

For the technical part. Use OpenCV with background subtraction and a blob detector.

Regards,

Joachim

José Gabriel Pereira Tavares I appreciate your suggestion about Yolo. I’ll definitely look into Kaggle, Huggingface, and Paperswithcode for smoke detection models.

Joachim Pimiskern The system, as it is currently designed, cannot operate in both directions simultaneously. This is a significant flaw, especially if a situation arises where smoke or fire is detected in both zones at the same time. Although such a situation has not yet occurred, we need to be proactive in finding a solution to avoid this potential problem. Unfortunately, the system was designed to operate in only one direction at a time. I am actively seeking a solution to this issue. I’m curious to know what others are using in this case for smoke and heat exhaust ventilation systems. Thank you very much for your responses.