What a pool equipment photo can show

Object Detection & Classification

The first stage of analysis identifies and localises individual equipment components within the uploaded image. The system can look for common pool equipment such as pumps, filters, chlorinators, heaters, valves, unions, pressure gauges, electrical enclosures and plumbing fittings. Each detected component is bounded with a confidence score, and the spatial relationships between components are analysed to understand the system topology - which components are connected, the flow direction, and the overall system architecture.

• Multi-class object detection across 15+ pool equipment categories
• Confidence scoring for each visible item
• Spatial relationship analysis between connected components
• Orientation and mounting position assessment

Condition Assessment Engine

Once components are identified, a specialised condition assessment model evaluates each one for signs of degradation, damage, or impending failure. The assessment considers: Visual indicators: Corrosion patterns, discolouration, calcium deposits, UV degradation, cracking, deformation, moisture staining, and biological growth. Contextual analysis: Component age estimation from visual wear patterns, installation quality assessment, and compatibility checking between connected components. The system assigns severity ratings (Low/Medium/High priority) based on the potential consequence of the identified condition - a corroded union near electrical components rates higher than cosmetic discolouration on a filter tank.

Anomaly Detection

Beyond identifying known failure modes, the system employs anomaly detection to flag unusual conditions that may not match specific training categories but deviate from expected normal appearance. This may flag unusual layouts, non-standard fittings or early visible wear that should be checked by a technician.

Recommendation Generation

Diagnostic findings are synthesised into actionable recommendations using a knowledge base of pool equipment maintenance best practices. Each recommendation includes: • Urgency classification (immediate attention, monitor, routine maintenance) • Specific remediation steps appropriate for the identified condition • Likely repair category, where enough information is visible • Consequence of inaction (what happens if the issue is ignored) Recommendations are prioritised by risk - safety-critical issues (electrical, structural) always surface first, followed by performance-degrading conditions, then cosmetic concerns.

Model Architecture & Training

The underlying vision system uses a multi-stage architecture combining feature extraction, object detection, and classification networks. The model is trained on a curated dataset of pool equipment images captured across diverse conditions, lighting environments, and equipment generations. Feedback from confirmed jobs can be used to improve future photo checks, where suitable.