[Peter]

One of the best things you can do to improve biogas production and improve stability it is feed a consistent amount of energy to you the digester every day. When the digester gets a really rich meal with lot of digestible solids one day then a small meal with a lot of water the next it is hard for the digester to grow a healthy population of bugs and impossible to create steady gas. Would recommend tracking how much digestible energy is being fed to the digester every day. If the amount of gas is not equal to the energy being fed it is time to look for possible causes of inhibition.

For more information on what information to collect on digester to keep biogas production high and stable check out our article: Top 10 Things to Make Your Digester Biology Happy.

https://azuraassociates.com/digester-biology-top-10/

azuraassociates.com

Top 10 Things to Make Your Digester Biology Happy – AZURA

Top 10 Things to Make Your Digester Biology Happy – AZURA

[Hatem]

While this advice is sound in principle, it’s important to acknowledge that, in a real-world operational context, manually controlling all the variables influencing digester performance is extremely complex—if not impossible. This is where AI becomes essential. By analyzing large volumes of operational data, AI can help identify the key parameters that require attention and guide operators on what to adjust, and when, to maintain optimal conditions.

I recommend tracking the amount of digestible energy fed to the digester daily and comparing it to the biogas output. If gas production doesn’t align with the expected energy input, it may indicate a need to investigate potential inhibition.
Thanks

[Peter]

Would like to push back on the need for AI tools. We work with several the food waste digesters and manure/food waste co-digestion facilities that receive a complex feedstock. These plants are able to predict daily gas production to within 5% with only basic lab testing and excel based tools used by the operators on site.

AI tools still need to get feedstock characteristics and digester health indicators from somewhere and conventional chemical tests currently produce more reliable results than inline spectroscopy-based tools or industry averages pulled from a database.

Additionally, a facility’s knowledge of the local feedstock landscape, how to handle them, and how to best operate their digester is part of the facility owners IP that can help them hold a competitive advantage. Plant owners might not want their operational lessons and data to be used to train the competition.

Finally, the currently available inline sensing technologies cannot screen for all possible toxins and contaminants in feedstock. Without a perfect sensor, plants still need well trained and knowledgeable operators and regular lab testing to make informed decisions about receiving new feedstocks. Since you need a skilled operator checking the AI’s work – I am not sure how investing in such a tool is expected to justify the considerable cost.

Would be happy to be educated about cases where AI tools have generated additional revenue for digesters that are performing well.

Thanks

[Hatem]

Thank you for sharing your perspective. I respectfully disagree with your assessment, as it underestimates both the current capabilities and the future potential of AI tools in anaerobic digestion (AD) operations. Your comments reflect a conventional, lab-centered view of process management, which does not scale effectively to the operational complexities of commercial facilities.

First, scaling up from lab tools and Excel-based forecasting to full-scale plant operations is not a linear process—they are fundamentally different worlds. A commercial digester processing tons of heterogeneous feedstock per day, with fluctuating input composition, operational disturbances, and dynamic energy markets, demands a level of system-wide integration and predictive capacity that simply cannot be met with spreadsheets and manual inputs alone. What works in a stable academic or pilot setting does not generalize to the volatile conditions of real-world operations.

Second, the AD business does not operate like traditional chemical processing plants where one-size-fits-all lab routines suffice. We’re dealing with living systems that vary daily in microbial dynamics, temperature responses, substrate synergy, and inhibition potential. AI excels at learning patterns in such complex, non-linear systems—especially when combined with continuous or semi-continuous data acquisition.

Regarding your valid point that AI needs data from conventional tests—yes, absolutely. But the value of AI is in how it connects, interprets, and predicts using these data streams. It doesn’t replace chemistry or expertise—it augments them.

On the concern that plant owners might not want to share their data for training purposes: that is a business model and data governance issue, not a limitation of the technology. Models can be customized per facility and trained solely on their own data to protect IP while still yielding substantial insights. AI is not just for optimizing yield; it’s about risk mitigation, early fault detection, and predictive maintenance—key to uptime and revenue protection.

Also, the argument about sensing technologies not being perfect is fair but misses the point: AI doesn’t depend on a single sensor. Its strength is in integrating multiple imperfect signals to yield a robust forecast or recommendation. It’s the same principle we trust in aviation, finance, and weather forecasting.

Finally, and critically:

1. Building a structured, historical data repository is essential for consistent operational knowledge transfer. This ensures that insight doesn’t walk out the door when a staff member leaves. AI or not, data discipline adds resilience to your operation.

2. A historical database supports standardization and benchmarking. With structured records, operators can compare performance across years, seasons, or feedstock shifts, and refine their strategy accordingly. It’s not only about AI; it’s about professionalizing operations and making informed decisions with traceability.

In short, AI tools are not a silver bullet, but to dismiss them based on lab-scale practices and assumptions about operator skill levels is to ignore the operational reality and business potential of modern digesters. We’re not replacing people—we’re giving them better tools to compete in a changing energy and waste landscape.

Would be glad to continue the conversation and share some concrete examples where data-driven models have made a measurable difference in well-performing plants.

Note:
I’ve attached a quick example from one of our applications, LimusAI, that shows how AI can precisely identify which features have the greatest impact on biogas production at a given site. This level of insight empowers operators with targeted, data-backed decisions that go far beyond what’s possible with manual tools alone.

[Peter]

Thank you for the detailed response @gazeol-renouvelable_hatem.

There are a lot of things we agree on.

1. Whole heartedly agree that “a structured, historical data repository is essential for consistent operational knowledge transfer.” Data discipline is critical and when AI tools help operators get there that is a good thing.

2. Agree that “a historical database supports standardization and benchmarking. With structured records, operators can compare performance across years, seasons, or feedstock shifts, and refine their strategy accordingly.”

Could even add that this data makes the facility more valuable and scalable because it grants a greater understanding of how process changes can affect digester performance.

This data discipline and gathering a well contextualized multiparameter data set over years would be valuable to many operating digesters and is valuable to the full-scale digesters and fleets that we work with to help implement robust monitoring protocols.

In cases of aviation, finance, and weather forecasting there are many well instrumented parameters monitoring a very large number of analysed processes. For existing biogas plants with largely manual data sets frequently only tracking a few parameters like pH, temperature, and tonnes fed per day, there may not be a signal to train a predictive model on.

Once a plant has been collecting a robust data set for several years, I think the potential for using AI tool to learn insights, potentially optimize, and assess future possibilities is exciting.

I remain skeptical about the necessity of AI tools for digester operation. Many full-scale digesters run well with just the detailed process monitoring that you describe, and without AI tools. Think that owners should carefully consider the cost and value of such tools after they have implemented the process monitoring and data handling practices needed to reduce their risk.