Skip to content
Koidra
Book a demo

Root cause analysis in hours, with AI

Customer story Maintenance Process Optimization Biomass energy

A fishbone diagram of a failure's possible causes, under machine, method, material, people, measurement and environment, with threads connecting evidence on several bones to the root cause.

A plant runs a root cause analysis whenever performance falls short: a machine that keeps failing, a yield that stays flat, an OEE (overall equipment effectiveness) that stays low. In most plants, that analysis is slow and often stops short of the real cause.

Why root cause analyses fall short

  • They take long. Gathering the records, talking to each shift and agreeing on a cause takes days or weeks, so most problems are never analyzed at all.
  • The records are incomplete. A work order closes with a tick and no measurement, and the readings that would explain a failure sit in another system.
  • Management decisions go unnamed. People write the report, and they rarely name the decisions made above them, so the analysis stops at the operator or the part.
  • Nobody holds the whole picture. Connecting the dots takes the machine’s history, the process data, the procedures and the lessons of past failures, and one person rarely knows all of them.

How Koidra’s AI finds the cause

Koidra’s AI reasons over the plant’s knowledge graph: its assets and how they connect, every metric over time, the work orders and procedures, and the lessons of past analyses. Physics models of energy, heat and mass keep its explanations to what the plant can physically do.

The knowledge graph under the digital twinAreas and lines hold the assets, each with its specs and sub-assets, the assets upstream that feed it, the parts fitted to it and its work orders. Metrics hold all the numeric data over time: an asset's operating conditions, from sensors, setpoints, operator entries and formulas, and the performance of each production batch, such as yield, energy per tonne and quality. The whole graph is the backbone of Koidra's physics-informed digital twin, which grounds the AI agent's root causes and recommendations.groundsKnowledge graphholdfeedfitted toonoperating conditionsperformance of eachAreas and linesAssetsspecs, sub-assetsUpstream assetsPartsWork ordersMetricsall numeric data over timeProduction batchyield, energy per tonne,qualityDigital twinphysics-informed modelsThe AI agentroot causes, recommendations
An asset's specs and the assets that feed it set its operating conditions. Metrics record those conditions and each batch's performance over time, and the AI reasons along that chain.

It writes the first report with the evidence for every cause, and an action plan in which every action has an owner. Many actions are changes it makes in Koidra right away, such as a new metric, a better alarm, an updated procedure or a work order.

At Ayo Biomass

Ayo Biomass makes wood pellets in Vietnam, about 400,000 tonnes a year from its own two plants. Its root cause analyses now come from Koidra’s AI. The team gives feedback on the first report, and a validated action plan is ready within hours, where it used to take days or weeks. The reports are also more precise than before.

From the AI's first report to an action plan, in hoursKoidra's AI writes the first report. The plant's team gives feedback and the AI revises it, then the team validates the action plan. Actions that live in Koidra, such as a new metric, a better alarm, a procedure step or a work order, are applied at once.First reportby the AIFeedbackand revisionAction planvalidatedAppliedin Koidra
The team checks the AI's report, and the actions that live in Koidra are applied as soon as the plan is agreed.

A sample report

Three rollers failed on one pellet mill in 18 days. The AI traced them to worn roller shells, missing grease and heavy load, and found a warning in the data one to two hours before each burnout. The report is shown with Ayo Biomass’s permission, with confidential information anonymized.

The first page of the sample report: its title, four headline facts, the summary and the start of the timeline.

Read the full report (PDF, 8 pages)

Root cause analysis is part of the AI Consultant in Process Optimization Pro.

See Koidra on your own plant data

Bring a week of readings from one line. We will show you what the platform sees, what it would have flagged, and what it would have scheduled.

Book a demo