【Original Tool】Afraid to let AI calculate wall thickness for fear of damaging the sewing machine? I hand-crafted a “Zero Hallucinations” GB150 review external brain

Thread Content

Hello, colleagues and seniors. I originally worked in process equipment, but in recent years my focus has shifted to big data development. Recently, large models (like ChatGPT) have become very popular, and many people in our industry groups are discussing whether these tools can be used to help us prepare calculation reports or review drawings As someone in the industry who knows a bit about the underlying code, I usually advise against it outright: anyone who dares to sign off on results generated by general-purpose large models is simply playing with their career. As we all know, the tolerance for errors in our industry is 0. If you ask in a general-purpose large model, \"According to GB150, what is the allowable stress at 250°C for Q345R steel plates with a thickness of 38 mm?\" ” It’s highly likely that it will seriously make up some random figure of 150 MPa for you. It simply can’t determine the thickness grades, nor can it figure out the condition of the material. This kind of “numeric illusion” is at most a joke in other contexts; but when it comes to pressure vessel design, it either results in excessive weight and criticism, or it causes the equipment to explode and crush something like a sewing machine. Current AI systems simply cannot understand those complicated cross-table formats and temperature gradients specified in our GB150.2 standard. Since I really couldn’t stand such nonsense, and also to address the problem of making mistakes due to difficulty in reading the minimum temperature values specified in manuals and documents, I took some time to thoroughly restructure the data in GB150. I happened to develop an interface-equipped GB150 intelligent review system. I’m posting it today; please, experts, give your feedback to see how it works. This system avoids any sort of mysticism and focuses on three key functions: 1. An \"accurate lookup\" function that never generates false information – I have converted all the standard material tables into structured data. For example, if you enter “20MnMo steel forgings used in low-temperature containers”, the system will not waste words and will immediately show a red light to indicate that the minimum operating temperature for this material is 0°C, so it is not suitable! It also automatically recommends alternative materials such as 09MnNiD forgings. 2. Auxiliary review (to prevent minor mistakes): Enter the design parameters submitted by others; for example, if a certain cylinder uses 12Cr1MoVR and a temperature of 575°C, the designer may have specified a stress level of 50MPa. The system will automatically identify the original table, warn you of an incorrect value – the correct value should be 41 MPa – and indicate a risk of high-temperature creep. This saves a lot of effort for brothers who usually do proofreading. 3. The most crucial point: 100% traceability in compliance. For those dealing with our equipment, acting without a regulatory basis is nothing but recklessness. For all the formula derivations and data in this system, there is a “Trace Origin” button below them. Click once, and the relevant clauses from GB150 in their original form along with screenshots of the original tables will be displayed immediately. It’s not very convenient to post multiple images and long videos on the forum, so I’ve compiled the video demonstrations of the actual operation (including how to calculate wall thickness accurately and how to trace the origins) as well as the underlying technical logic into a WeChat post. Colleagues who are interested can click on the link below to watch the live demonstration video: 👉 Live Demonstration and Technical Breakdown: https://mp.weixin.qq.com/s/jvdNkm5NwNB2v-Bt_f8APA(The videos on that link are truly impressive; it’s recommended that everyone take a look to see what can be achieved when large models are combined with robust industrial standards.) At present, this underlying dataset and review pipeline have been successfully implemented. I’m going to take it out for everyone to test. If there are colleagues who work on the design and review of static equipment in the field, or those interested in the combination of industrial software and AI, feel free to leave a comment below the post or add me on WeChat by reading the article. I’m providing a few test accounts for everyone to use, so that you can test its limits with those extreme operating conditions you have at hand. I welcome criticism and suggestions from my colleagues and seniors!