From a Single Question to a Root Cause Investigation
For years, industrial AI has been associated with large projects.
Build an AI agent.
Connect it to your historian.
Integrate alarms.
Map your asset hierarchy.
Connect dashboards.
Configure analytics.
Only then can you start asking questions.
One thing I’ve enjoyed about the TDengine Historian is that none of that is required.
The platform already includes built in AI and an embedded AI agent designed for industrial workflows. Instead of spending weeks building an AI stack, you simply open the AI workspace and start asking questions about your operations.
I wanted to see how far that experience could go.
Starting with a Simple Question
Rather than beginning with SQL queries or dashboards, I asked one question.
“Write me a report on my pharmaceutical manufacturing cluster.”
Within seconds, the TDengine Historian began building the report.
Behind the scenes it explored the asset hierarchy, identified every manufacturing site, gathered historian data from formulation tanks, fill and finish lines, packaging equipment, and cleanroom systems, reviewed alarms and events, compared performance across facilities, and assembled everything into a structured operational report.
Instead of looking at a few trends, I had a complete summary covering three manufacturing sites, twelve monitored assets, and tens of thousands of data points.
The report highlighted something that immediately stood out.
The Boston manufacturing site was consistently performing worse than the other two facilities.
Finding the Problem
The report identified several differences between Boston and the Dublin and Singapore sites.
The Boston fill line was operating with a 35% higher reject rate.
It also showed:
- Higher fill volume deviation
- Packaging line microstops
- Increased vision reject rates
- Repeated cleanroom differential pressure alarms
Even more interesting, the report showed that many of these issues improved after an intervention around June 28 but never fully returned to the baseline achieved by the other manufacturing sites.
At this point the report had already saved a significant amount of investigation time.
Instead of manually opening dashboards, comparing sites, exporting historian data, and reviewing alarms individually, everything was summarized in one place.
But the investigation wasn’t finished.
Continuing the Investigation
Rather than opening another tool, I simply continued the conversation.
“The BOS fill and finish line has a persistent rejection problem. Please write me some recommendations on how I can solve this.”
This is where the built in AI became much more interesting.
Instead of producing generic troubleshooting advice, it returned to the historian and performed a much deeper analysis.
It compared seven days of production data across all three manufacturing sites.
It calculated the relationship between reject rate and fill volume deviation.
It identified process changes during the reporting period.
It compared BOS against the healthier Dublin and Singapore sites to establish a performance baseline.
Within seconds it identified something that would normally require much more manual analysis.
The reject rate and fill volume deviation had a correlation coefficient of 0.928.
In other words, nearly all of the increase in reject rate could be explained by poor fill volume control.
The AI also detected a clear process change around June 28 where both metrics improved significantly, suggesting maintenance or calibration work had already taken place.
The improvement was real.
But it wasn’t enough.
Boston still remained well above the performance of the other two sites.
Recommendations Based on Operational Data
Rather than providing a generic checklist, the recommendations were directly tied to the operational evidence.
The highest priority actions included:
- Recalibrating the fill volume dosing system to match the performance of the Dublin and Singapore sites.
- Inspecting fill nozzles and dispensing valves for wear, clogging, or mechanical issues.
- Reviewing exactly what maintenance occurred during the June 28 intervention to understand why it only partially resolved the problem.
- Verifying fill volume sensor calibration against laboratory measurements.
- Comparing equipment configuration and maintenance history between all three manufacturing sites.
- Creating real time monitoring and alerting for reject rate and fill volume deviation so issues could be detected earlier.
Every recommendation was supported by historian data, production metrics, or statistical analysis.
Nothing was simply guessed.
This Is How Engineers Actually Work
What I found most interesting wasn’t the report itself.
It was the workflow.
The conversation naturally evolved from:
“Write me a report.”
to
“Why is this happening?”
to
“What should I do next?”
That’s much closer to how engineers investigate problems every day.
One answer leads to another question.
Another question leads to another analysis.
Eventually, the investigation converges on a likely root cause.
Instead of switching between historians, dashboards, spreadsheets, alarm systems, and analytics tools, the entire investigation stayed inside a single interface.
More Than a Chatbot
One thing that’s easy to miss when people talk about industrial AI is that the value isn’t simply asking questions.
The value comes from everything happening behind the scenes.
Within the TDengine Historian, the built in AI and embedded agent can work directly with:
- Historian data
- Asset hierarchies
- Events and alarms
- Dashboards
- Advanced analytics
- Reports
- Semantic reasoning
- Follow up investigations
Because these capabilities are already part of the platform, there’s nothing additional to deploy before getting started.
You simply begin asking questions.
For organizations that want to bring their own AI, the TDengine Historian also supports MCP, allowing external AI agents to securely access the same operational context.
But for many operators and engineers, the built in experience is enough.
Open the AI workspace.
Ask a question.
Continue the investigation.
Final Thoughts
I think this is where industrial AI becomes genuinely useful.
Not because it replaces engineers.
Not because it writes reports.
But because it shortens the time between identifying a problem and understanding why it’s happening.
TDengine Historian doesn’t just answer questions.
It helps engineers investigate, compare sites, analyze operational data, and uncover root causes using capabilities that are already built into the platform.
For industrial teams, that means less time collecting information and more time solving the problems that actually matter.
