Leveraging AI for Hydrogen Projects

[After the last newsletter, received some great feedback and questions from several readers about the article on assessing AI for a hydrogen microgrid project. To provide some additional context and details, this post is a scrubbed version of the AI section of my final report for the project. Hopefully it's of some interest to others who may be charting these waters.]

AI Overview

AI methodologies mindmap (image credit: Dr. Milan Milanovic)

There are a wide range of AI methodologies as illustrated above. Some of these AI types have been under development for many years (e.g., expert systems, speech recognition, game playing, etc.).

Impressive advances have been achieved in deep learning neural networks methods, particularly those using large language models (LLM) that incorporate enormous amounts of public domain data from the internet and other sources for training the model (e.g., AI chatbots). AI chatbots produce human-like responses based on user prompts.

Small language models (SLM) were also initially considered early in the project but ultimately not used. SLMs require identifying and labeling selected source data for training by subject matter experts. Since published codes and standards are copyrighted, there are unresolved issues about how to use SLM-based models without infringing on the associated intellectual property rights.

The most widely available and heavily used general purpose LLM-based AI chatbots in mid-2024 included OpenAI’s ChatGPT, Google’s Gemini, and Microsoft’s Copilot. As of early 2026, Anthropic Claude and Perplexity have now become top ranked general purpose options as well.

The interim report in mid-2024 contained responses and assessments of ChatGPT, Gemini, and Copilot relevant to hydrogen system safety. Since Copilot incorporates OpenAI models and is now more commonly used within the Microsoft ecosystem of products, it was not included in the most recent assessment.

 

Limitations and Constraints

 

There are a variety of limitations and constraints to keep in mind when using AI chatbots, including: prompts, responses, data sources, accuracy, and pace of evolution.

The responses from AI chatbots are highly influenced by the wording of the prompt provided by the user (aka, “prompt engineering”). Contextual information and perspective in the prompt helps to refine the response. Responses also change based on the conversation thread, timing of the prompt, and other parameters.

LLM models are trained with massive amounts of data available from the internet and other potential undisclosed sources. Confidential or proprietary data is not accessible unless provided by the user in an uploaded document or via app usage permissions. Most hydrogen codes and standards represent unresolved intellectual property issues regarding use as data sources as previously mentioned.

Generated responses are an amalgam of legitimate reference material, extrapolated information, and sometimes nonexistent sources (aka, “hallucinations”). Verification of responses by independent means should be done to the greatest extent possible via an internet search and review of any provided citations.

All of the AI Chatbots tested are in continuous development with new capabilities released frequently. As a result of their rapidly evolving performance and capabilities, any assessment is only valid within a limited timeframe.

 

Summary of AI Results

 

The AI chatbots were each given specific prompts on the following topics: describing how they work; hydrogen codes and standards; hazards analysis; summarizing a safety document; and safety training.

Initial Assessment in Mid-2024

 

Some specific areas where the individual AI chatbots performed better relative to each other in the mid-2024 assessment:

• ChatGPT (OpenAI) provided the best response on the topic of codes and standards; good response with details on hydrogen hazards analysis; and the best summary of types of training (but without details on sources or providers)
• Gemini (Google) provided a good response and formatting on hydrogen hazards analysis
• Copilot (Microsoft) provided the most detailed response summarizing a selected safety document, but the source document’s table of contents provided a much better overview

None of the AI chatbots were consistently better on all prompts, and none of them provided comprehensive information that could be used without supplemental methods and verification. Personal domain experience, benchmarking, and standard internet searches provided more comprehensive and verified hydrogen safety information compared to the AI chatbots.

For individuals or organizations inexperienced with hydrogen systems and related safety considerations, the AI chatbots may be useful as an initial starting point. However, they cannot be relied upon to provide comprehensive accurate information on hydrogen systems safety. They may also be useful as a complementary tool to use in conjunction with other methods such as internet searches, benchmarking, etc.

Copyright infringement and other intellectual property issues are a significant barrier to using AI tools to extract relevant data from codes and standards for a specific hydrogen system. Although none of them appear to include this data in their general responses, they all have the capability to upload or otherwise access user-provided documents that could contain protected or proprietary material.

One area where the AI chatbots currently exhibit very promising capabilities is text generation of new documents (e.g., summaries, checklists, outlining, etc.). Using these tools to create a draft for further refinement by an appropriately skilled person could significantly reduce the preparation time and potentially identify items that may not have been otherwise considered. An example prompt and response for a hydrogen microgrid system checklist are given in the final section that illustrates this capability using ChatGPT.

 

Updated Assessment in Early-2026

 

The responses from ChatGPT and Gemini to the same user prompts 18 months after the initial assessment illustrates how quickly these AI tools have evolved. The quality, details, and comprehensiveness of the newer responses is greatly improved.

While all the limits and constraints previously discussed still apply (along with the recommended mitigations), there is a remarkable increase in the utility of these AI tools. Whereas the initial assessment indicated they would be useful as a starting point for those with limited hydrogen experience, they are now a valuable augmentation even for subject matter experts.

Another interesting observation is that it is more difficult to discern specific areas where one tool is superior to the other. Instead, each performs quite well in its own right while also providing different perspectives and some unique supplemental content. In fact, an effective approach would be to use the same prompts with two or more of these AI tools and integrate the responses (e.g. Gemini, ChatGPT, Claude, Perplexity, etc.).

A few other AI themes are currently emerging that are worth considering. First, the combination of an expert human augmented with an AI “assistant” can produce better results than either can produce alone. Second, more specialized AI tools for specific use cases are rapidly growing and provide further improvements in application-specific domains (e.g., hydrogen safety). Finally, the growth of more advanced AI agents that proactively perform user-specified tasks rather than requiring human prompts can vastly improve productivity.

 

Checklist Example

 

Prompt: Create a checklist for starting and operating a hydrogen microgrid system that electrolyzes water using input power and water supply, compresses the resulting hydrogen gas, stores it in a compressed hydrogen pressure vessel, and feeds the hydrogen to a fuel cell stack when electrical power is needed.

 

2026-Jan-14 ChatGPT 4o response:

"Below is a practical, end-to-end checklist for starting, operating, and maintaining a hydrogen microgrid system that uses electrolysis, compression, storage, and a fuel cell stack.

It is structured so it can be used by operators, engineers, or commissioning teams, moving from planning → startup → normal operation → shutdown → safety & maintenance.

 

1. Pre-Project & Design Checklist (Before Installation)

 

System Architecture

☐ Define power sources for electrolysis (grid, solar, wind, backup generator)

☐ Define electrical loads and peak/average power requirements

☐ Select operating mode:

☐ Grid-connected

☐ Islanded

☐ Hybrid

☐ Establish hydrogen production vs. consumption balance

☐ Define storage autonomy (hours/days of backup)

 

Component Selection

☐ Electrolyzer type selected (PEM / alkaline / solid oxide)

☐ Hydrogen compressor type selected (mechanical / electrochemical)

☐ Storage vessel type selected (Type I–IV pressure vessel)

☐ Fuel cell type selected (PEMFC / SOFC)

☐ Power electronics selected (DC/DC, inverter, rectifier)

☐ Control system / PLC specified

 

Codes, Standards & Permitting

☐ Compliance with local electrical codes (e.g., NEC, IEC)

☐ Compliance with hydrogen standards (e.g., NFPA 2, ISO 19880)

☐ Pressure vessel certification (ASME / PED)

☐ Environmental permits (water use, venting)

☐ Fire marshal approval

☐ Emergency response plan documented

 

2. Installation & Commissioning Checklist

 

Site Preparation

☐ Adequate ventilation for hydrogen areas

☐ Classified electrical zones identified (hazardous area rating)

☐ Hydrogen leak detection installed and tested

☐ Fire suppression systems installed

☐ Physical barriers and signage installed

 

Water Supply System

☐ Feedwater quality verified (conductivity, purity)

☐ Water treatment system installed (RO / DI if required)

☐ Flow rate meets electrolyzer requirements

☐ Drain and purge lines installed

 

Electrical Systems

☐ Input power verified (voltage, frequency, phase)

☐ Grounding and bonding completed

☐ Emergency stop circuits tested

☐ Power isolation and lockout/tagout verified

 

Hydrogen System

☐ Piping pressure-tested

☐ Hydrogen-rated materials verified

☐ Check valves installed and oriented correctly

☐ Pressure relief devices installed and set

☐ Vent stacks routed to safe discharge locations

 

3. Startup Checklist (Initial System Energization)

 

Pre-Startup Safety Checks

☐ No hydrogen leaks detected

☐ All valves in correct startup position

☐ Storage vessel pressure within allowable limits

☐ Compressor oil/cooling systems operational

☐ Fuel cell stack isolated from hydrogen supply

☐ Control system alarms enabled

 

Electrolyzer Startup

☐ Water supply active and within spec

☐ Electrolyzer warmed up (if applicable)

☐ Power ramped gradually to operating level

☐ Hydrogen purity verified

☐ Oxygen venting confirmed

☐ Electrolyzer operating within temperature limits

 

Compression & Storage

☐ Compressor started unloaded

☐ Compression stages ramped gradually

☐ Intercoolers and aftercoolers functioning

☐ Storage vessel pressure increase monitored

☐ Automatic shutdown thresholds verified

 

4. Normal Operating Checklist

 

Electrolysis Operation

☐ Power input within design range

☐ Water consumption monitored

☐ Stack voltage and current stable

☐ Hydrogen production rate within expected range

☐ Oxygen venting unobstructed

 

Hydrogen Compression & Storage

☐ Compressor vibration within limits

☐ Storage pressure below maximum allowable working pressure (MAWP)

☐ Temperature within design limits

☐ No abnormal pressure decay (leaks)

☐ Storage state-of-charge logged

 

Fuel Cell Operation (Power Demand Mode)

☐ Hydrogen supply pressure within fuel cell specs

☐ Fuel cell purging sequence completed

☐ Stack temperature stabilized

☐ Power ramped gradually

☐ DC output stable before inverter connection

☐ Electrical load matched properly

 

Electrical Power Management

☐ Inverter synchronized (if grid-connected)

☐ Load sharing logic functioning

☐ Battery (if present) operating correctly

☐ Power quality (voltage, frequency, THD) within limits

 

5. Shutdown Checklist

 

Planned Shutdown

☐ Reduce electrical load gradually

☐ Ramp down fuel cell power

☐ Close hydrogen supply to fuel cell

☐ Purge fuel cell with inert gas or air (per OEM)

☐ Stop electrolyzer production

☐ Depressurize lines if required

☐ Secure compressors and auxiliary systems

 

Emergency Shutdown

☐ Activate E-stop

☐ Hydrogen supply isolated automatically

☐ Venting systems activated

☐ Electrical isolation engaged

☐ Area evacuated if hydrogen alarm triggers

 

6. Safety & Monitoring Checklist (Continuous)

 

Hydrogen Safety

☐ Leak detectors calibrated and operational

☐ Ventilation fans running

☐ Pressure relief devices unobstructed

☐ Flame detectors operational (if installed)

 

System Monitoring

☐ Pressure, temperature, and flow logged

☐ Stack health monitored (electrolyzer & fuel cell)

☐ Alarm thresholds reviewed periodically

☐ Remote monitoring functional

 

7. Maintenance Checklist

 

Daily / Weekly

☐ Visual inspection of piping and vessels

☐ Check for abnormal noise or vibration

☐ Review alarm logs

 

Monthly

☐ Compressor service checks

☐ Leak detection system test

☐ Sensor calibration spot checks

 

Annual

☐ Pressure vessel inspection

☐ Full electrical inspection

☐ Electrolyzer stack performance evaluation

☐ Fuel cell stack degradation assessment

☐ Emergency drills conducted

 

8. Documentation & Training Checklist

☐ Standard Operating Procedures (SOPs) documented

☐ Emergency response procedures posted

☐ Operator training completed

☐ Maintenance manuals accessible

☐ System diagrams updated (P&ID, single-line)

 

If you want, I can:

• Convert this into a one-page operational checklist
• Tailor it for PEM vs alkaline electrolyzers
• Adapt it for remote/off-grid operation
• Create a failure modes & troubleshooting checklist

Just tell me how detailed or specialized you want it."

Author Bio

Matt Moran is the Managing Member at Moran Innovation LLC, and previous Managing Partner at Isotherm Energy. He's been developing power and propulsion systems for more than 40 years; and first-of-a-kind gas, slush, and liquid hydrogen systems since the mid-1980s. Matt was also the Sector Manager for Energy & Materials in his final position at NASA where he worked for 31 years. He's been a cofounder in seven technology-based startups; and provided R&D, engineering, and innovation consulting to several hundred organizations. Matt has three patents and more than 50 publications including his online Cryogenic Fluid Management guide and Decarbonizing Mobility with Liquid Hydrogen SAE report. He has created and taught liquid hydrogen courses, webinars, and workshops to global audiences.