Status and Development Directions of Data Architecture for Intelligent Assessment of the Technical Condition of Mobile Machines of Agro-Industrial Enterprises

This article provides a critical survey and a practical method for assembling sensor datasets that support intelligent assessment of the technical condition of mobile agricultural machines. The survey maps available sources and shows that domain-relevant datasets are scarce, fragmented, and rarely documented with stable collection procedures. To address these gaps, the paper proposes a coherent “global to detail” pathway that links fleet decisions, field acquisition, and analytics into a reproducible blueprint. The pathway begins with fleet-level prioritization that identifies machine classes and critical components, outlines typical failure modes and observable symptoms, and balances severity, likelihood, detectability, seasonal load, safety, and downtime costs. The result is a prioritized roadmap for sensor installation and a compact measurement plan specifying signals, installation points, sampling frequencies, and event time- stamping. Data collection and management follow an offline-first architecture suitable for unstable connectivity in fields. Sensors feed an edge acquisition unit with unified time derived from global navigation satellite systems and the network time protocol; measurements are buffered locally and delivered with store-and-forward when links reappear, using lightweight message transports such as message queue telemetry and secure hypertext transfer. In the central environment, data land in a time-series repository or data lake governed by a metadata catalog that records dataset passports, provenance, versions, privacy and access rules, and licensing terms. A preprocessing layer performs extraction, transformation and loading, cleans artifacts, harmonizes time, and segments streams into versioned releases for model development. Analytics follow the functional reference defined by the ISO 13374 for condition monitoring: acquisition, manipulation, state detection, health assessment, prognostics with remaining useful life estimation, and advisory generation with integration into computerized maintenance management systems. Each dataset release carries a minimal passport and acceptance gates that verify regime coverage, synchronization accuracy, limits on missing data, signal-to-noise quality, class balance, and explicit lineage with semantic versioning. The survey also found no sources providing pinned computational environments that guarantee full reproducibility; future publications should include dependency lock files or container images and concise replication guides.