Public technical brief
Engineering intelligence.
Clearly explained.
Understand how ARG-BLC1 supports industrial chiller analysis, predictive insight and informed engineering decisions. Explore its capabilities and the evidence required to evaluate them.
Public capability overview. Proprietary implementation details are not published.
Six-page partner brief in English. Opens in a new tab for printing or saving.
Explore the brief
Select a chapterAn engineering framework for understanding chiller behaviour.
Temperature, pressure, flow and power become useful evidence when their definitions, quality and operating context are understood together.
The system supports engineering judgement. A departure from a reference begins an investigation; it does not establish a failed component.
Engineering context makes changing behaviour interpretable.
Supporting measurements and equipment-specific engineering context help engineers examine changing behaviour. Anomaly indicators highlight observations for investigation; they do not independently establish a cause.
Reviewed outcomes can inform future improvements. A prediction error does not automatically become a confirmed equipment fault.
Predictive insight must be evaluated against measured outcomes.
Forecast evaluation compares predictions with relevant observations and an appropriate reference. Data coverage, operating conditions and the intended use determine how the result should be interpreted.
Future predictions have no measured accuracy score until matching observations are available. No error level or accuracy threshold is guaranteed.
Improvement depends on evidence and engineering review.
Research and reviewed operating outcomes can suggest improvements. Proposed changes require a clear benefit, relevant evaluation and an honest account of limitations before release is considered.
A reviewed proposal still needs human approval. Research discovery does not automatically retrain, release or promote a model.
Sensitive operations and changes require appropriate governance.
Controlled access, technical review and human authorization support responsible use. Operational deployment requires its own acceptance process and clearly assigned responsibilities.
Security configuration and operational acceptance require deployment-specific evidence. No certification or breach-resistance guarantee is claimed.
A pilot starts with agreed inputs and acceptance criteria.
Equipment identity, sensor and meter definitions, time-aligned history and a reviewed operating reference make comparisons interpretable. Maintenance evidence and agreed criteria support a fair pilot evaluation.
Equipment compatibility is profile-specific. Software tests and historical replay do not establish commissioned field performance.
Human oversight remains central to responsible use.
This is an engineering pilot and decision-support system. Anomaly indicators guide investigation, while fault findings and operational decisions require appropriate engineering evidence.
The public website has no plant connection and issues no equipment commands. It is not a certified safety protection system.
Conceptual illustration of the brief, not system architecture. Select with a click or tap. Use arrow keys to move between chapters. All seven sections are available below
Purpose
Industrial chillers communicate their operating performance through a continuous stream of data—temperature, pressure, flow, and power. ARG-BLC1 transforms these signals into meaningful engineering insights, enabling a deeper understanding of system behaviour and performance.
PROJECT CASSANDRA integrates engineering expertise with advanced machine intelligence within a structured, data-driven framework. By correlating operational measurements, identifying evolving performance patterns, and validating predictive models against real-world outcomes, it establishes a rigorous and intelligent approach to system analysis, performance optimisation, and predictive decision-making.
It is an engineering pilot and decision-support system. A detected departure from a reference is evidence for investigation, not proof that a component has failed.
Connected capabilities
Quality-aware measurements
Unit definitions, timestamps, gaps, missing values and physical plausibility are part of the analysis. Invalid or unsupported observations should remain visible rather than silently becoming normal operation.
Thermodynamic context
Temperatures, pressures, electrical input and flow are interpreted against an equipment-specific operating reference. Load and operating conditions matter when evaluating changing efficiency or component behaviour.
Explainable investigation
Analysis retains supporting measurements and alternative explanations across heat exchangers, compressors, refrigerant behaviour and sensors. Conclusions must distinguish an observation, a hypothesis and a confirmed finding.
Anomaly investigation
Unusual patterns can direct attention to observations that merit review. Their significance depends on equipment context and supporting evidence; an indicator alone does not establish a failed component.
Reviewed learning
Reviewed outcomes can inform candidate improvements. Later measured values can evaluate forecasting models. These are distinct forms of evidence; prediction errors are not automatically treated as verified equipment faults.
Forecasts that can be checked
Predictive capabilities are assessed using measurements that become available after a prediction is made. Evaluation must fit the equipment, intended use and available data.
| Evaluation principle | Why it matters |
|---|---|
| Measured outcomes | Relevant observations provide evidence of how a prediction performed. |
| Reference comparison | A suitable reference helps establish whether a forecast adds useful information. |
| Coverage and context | Data availability and operating conditions define how far the evaluation can be applied. |
What the scores mean
RMSE measures the typical scale of prediction error, with larger errors receiving more weight. For EER prediction, RMSE has the same dimensionless ratio basis as EER.
R² compares squared prediction error with the variation in the measured values. A negative result can be valid; R² is undefined when the measured target is constant. Scores need enough matched observations and must be read with data coverage and the reference result.
A controlled path to improvement
External research and reviewed operating experience can suggest improvements. Evidence and engineering judgement determine whether a proposed change is suitable for its intended use.
- Define the intended benefit and relevant limitations.
- Evaluate the proposal using representative evidence and agreed criteria.
- Review errors, regressions and unresolved uncertainties.
- Confirm whether the results support the proposed use.
- Require appropriate human authorization before operational release.
A proposal does not grant permission to update production. Research discovery does not automatically retrain or release a model, and a successful software evaluation does not establish field performance.
Security and control
Security and change governance support responsible use of engineering intelligence. Their effectiveness depends on deployment configuration, technical review and operational acceptance.
Controlled access
Sensitive operations require appropriate access controls and clearly assigned responsibilities. The public introduction website has no operational connection to private services.
Human review
Proposed changes require relevant evaluation and appropriate human authorization. A review decision does not by itself establish that a change is ready for operational deployment.
Deployment acceptance
Security configuration and suitability for the intended environment require separate assessment. Software capabilities are not evidence that every deployment is securely configured or independently tested.
Operational responsibility
Release decisions, maintenance and recovery arrangements should be defined and assessed for the intended environment. These remain operational responsibilities beyond this public overview.
What a pilot evaluation needs
| Requirement | Why it matters |
|---|---|
| Equipment identity | Model, cycle, refrigerant and relevant technical definitions establish whether an analysis profile is suitable. |
| Measurement definitions | Verified units, sensor meaning, fluid properties and electrical meter boundaries prevent incorrect calculations. |
| Time-aligned history | Known timestamps, sampling intervals, coverage and quality rules support a fair comparison. |
| Reviewed operating reference | A commissioning reference helps distinguish changing conditions from unexplained departures. |
| Maintenance evidence | Technician-reviewed events are needed to validate fault findings independently. |
| Agreed acceptance criteria | Forecast errors, false alarms, missed events, operating coverage and release responsibilities should be defined before evaluation. |
Scope and boundaries
- Equipment compatibility is profile-specific. Universal support is not claimed.
- Anomaly indicators are investigative hypotheses, not confirmed component failures.
- The system is not a certified safety protection system or a source of automatic equipment commands.
- Described capabilities do not establish a commissioned live deployment or independently verified field accuracy.
- Historical replay, simulation and software tests must remain distinct from live field validation.
- Public research discovery does not automatically retrain or release a production model.
This brief is a public capability overview. Proprietary implementation details are not published. It contains no customer data, plant telemetry or operational access details.
Return to PROJECT CASSANDRA