Measurements in one system
Automatically acquire analogue and digital signals from controllers and measuring devices. Monitor data delivery and display current values.
SOLUTIONS · PROCESS DATA
Process data systems
From measurements to predictions and decisions.
Acquire and analyse process data to monitor equipment condition. Further development includes failure prediction, assessment of impacts on connected equipment, and AI-assisted recommendations for process control.
01 / FACILITY DATA
A unified equipment operating history
The process data system combines measurements, events and operating modes. Technical metering data adds context by relating load and energy use to equipment performance and condition.
Facility operating history
Long-term storage of process parameters. Work with COMTRADE files and group measurements by equipment and facility groups.
From chart to event
Tables and charts of analogue and discrete signals. Find extrema and periods of rising or falling values within a selected time range.
Browser access
Web interface, mimic diagrams and reports. Metering, telecontrol and process analysis functions are selected as part of the solution.
Link to technical metering. Load profiles, currents, voltages and power help assess operating changes. Additional measurements are selected to diagnose specific defects at each facility. More about technical metering →
02 / PROJECT-SPECIFIC EXTENSION
Predictive diagnostics and failure prediction
The process data archive provides a foundation for specialised diagnostic models. Analytics scope and prediction horizon depend on available measurements, fault history and on-site model validation.
Detect deviations
Compare parameters under equivalent operating conditions and identify anomalies and persistent signs of deterioration.
Output: observed indicators and diagnostic priority.
Assess failure risk
Analyse trends in diagnostic indicators, operating hours and maintenance history. Estimate risk over a specified horizon.
Remaining useful life is reported as a range when data is sufficient and model accuracy has been validated.
Work priorities
Assess equipment condition alongside criticality, backup availability and the production impact of a shutdown.
Output: recommendations for inspections and maintenance planning.
Data needed for reliable diagnostics
- Electrical parameters: load profiles, currents, voltages, active and reactive power, and protection events, where available.
- Process parameters: temperature, vibration, pressure, flow, speed and equipment states, selected for the relevant failure modes.
- Context: operating modes, running hours, repairs, confirmed faults and service conditions.
- Quality: time synchronisation, units, quality flags, missing-data checks and mapping of signals to equipment.
Profiles at 15–60-minute intervals are suitable for load trends but usually insufficient to diagnose a specific mechanical defect. Sampling rate and delivery latency depend on the required warning time. Daily transfers support retrospective analytics; timely warnings require an appropriate data channel.
If data is insufficient, the system should flag an anomaly and request inspection. An exact failure date is not promised.
03 / EQUIPMENT DEPENDENCIES
Impact on other equipment
A dependency model combines power supply diagrams, equipment dependencies, shared resources, redundancy and constraints. It supports assessment of shutdown, load redistribution and fault propagation scenarios.
Example scenario: a cooling circuit pump. Energy use increases at comparable flow and head, alongside a change in temperature or vibration. These indicators call for investigation.
Equipment condition
The diagnostic model assesses deviations, possible causes and risk in the current operating mode.
Dependent consumers
The circuit diagram identifies affected units, potential reductions in cooling and performance constraints.
Backup and shared resources
Check standby pump readiness, the shared power supply section and permissible loading in the selected scenario.
Illustration of the decision logic. Causes and consequences must be supported by measurements, engineering rules and models specific to the facility.
Calculation basis. Dependencies are defined from documentation and verified by specialists. Complex scenarios may use a digital twin. The LLM uses calculation results to prepare explanations.
04 / DECISION SUPPORT
LLM-assisted recommendations for process control
A large language model connects diagnostic results with manufacturer instructions, procedures and maintenance history. Operators receive an explanation, possible actions and the rationale for choosing them.
Calculate
Identify fault indicators and estimate failure risk and impacts on connected equipment.
Explain
Compare results with approved documentation, find similar cases and generate recommendations with supporting references.
Decide and execute
The operator decides. Existing control algorithms execute authorised actions within active constraints and interlocks.
What a recommendation contains
- Situation
- Equipment, detected indicators, operating mode and data timestamp.
- Assessment
- Risk over the selected horizon, uncertainty and possible impacts on other equipment.
- Actions
- Required checks, permissible operating changes and conditions for applying them under the procedures.
- Evidence
- Source measurements, model output, instruction references and verified similar cases.
- Outcome
- The operator’s decision, actions taken and subsequent changes in condition.
Advisory mode. Recommendations appear at the operator workstation or in an agreed SCADA interface. The LLM has no direct access to actuators. Action validation, protection and interlocks remain in the established process control system; decisions and outcomes are logged.
05 / PHASED IMPLEMENTATION
From facility data to a validated model
Analytics can be developed on an existing process data system. Integration with technical metering, SCADA and maintenance systems depends on the project and available interfaces.
- SurveySelect critical equipment and failure modes; check measurements, archive quality and process dependencies.
- Pilot diagnosticsConfigure models for a selected equipment group. Assess warning lead time, missed detections and false alarms.
- RecommendationsConnect approved procedures and the LLM. Evaluate the validity of conclusions and usefulness of recommendations with operating staff.
- OperationIntroduce advisory mode, logging and quality control. Expand coverage after the pilot results are validated.
For enterprises in Kazakhstan, analytics hosting, access control, redundancy and data exchange are determined by facility requirements. Local LLM deployment is assessed when selecting the model and infrastructure.
International experience and methodological sources
- Siemens Senseye Predictive Maintenance ↗ — combining machine learning, generative AI and specialist experience to support maintenance.
- NIST: digital twins ↗ — condition monitoring, anomaly detection and system behaviour analysis.
- NIST AI 600-1: generative AI ↗ — checking the reliability, sources and limitations of language model outputs.
The sources describe technological approaches. The functions, integrations and quality targets of the Almaty Project solution are defined through the technical specification and testing.