TDengine

Global – October 28, 2026 -- Neither MQTT retained messages nor persistent sessions can guarantee complete production history, so industrial operators who need every measurement must add application-level buffering and a historian, according to an engineering analysis of the OASIS MQTT 5.0 specification.

Retained messages deliver the latest state to new subscribers

A retained message gives a subscriber the last retained value on a topic when it subscribes. In the example of a packaging line publishing to plant/line2/packer/status, a new maintenance display receives the retained RUNNING state without waiting for the next live publication, provided the message is still available and the subscription's retain-handling settings permit delivery.

Global – October 27, 2026 -- Industrial teams can locate missing historian records by tracing one test message across five boundaries: broker receipt, subscriber receipt, parsing, database write and query. The approach comes from troubleshooting guidance on TDengine's MQTT ingestion, which warns that a live broker feed or a connected publisher proves only part of the chain.

One tagged test message anchors the diagnosis

Engineers should publish a single message for one equipment tag through the approved production path, such as plant/line2/pump07/pressure. They should log the exact topic, payload bytes, broker timestamp, publisher identity, destination table and intended source timestamp. Each test needs a fresh timestamp, because identical timestamps can make a successful write indistinguishable from an earlier one.

Global – October 27, 2026 -- Even QoS 2's exactly-once delivery covers only one MQTT sender-receiver flow, not the subscriber's database transaction, so an MQTT broker cannot serve as an industrial historian. A broker moves messages between publishers and subscribers. A historian keeps time-indexed records that engineers can query, trend and aggregate later.

Broker retention mechanisms fall short of historical storage

A retained message is only the current message for a topic, and a new one normally replaces the prior state. A persistent session preserves subscriptions and eligible queued messages for an offline client. Session expiry, broker capacity, message expiry and administrative policy can all limit what remains.

Broker persistence protects broker state across restarts. It does not define table schemas, engineering timestamps, multi-year retention, SQL queries, aggregates or audit-ready change history.

Global – October 27, 2026 -- An MQTT message is only a topic plus a payload, so a usable historian record needs four additions: stable equipment identity, typed measurements, a trustworthy timestamp and a schema that survives device and site growth. Engineers should settle these before configuring ingestion into TDengine TSDB Enterprise. Skipping that step can produce technically successful writes that create tables that are hard to query, compare or maintain.

Global – October 27, 2026 -- TDengine is framing Industrial AI as decision support built on three layers (data, context and analysis) rather than as autonomous plant control, with accountability remaining with engineers and operators.

Industrial AI applies standard methods to physical operations

The approach uses machine learning, computer vision and generative AI on data from equipment, sensors, control systems, operational records and engineering documents. What makes an application industrial is the work it supports: pumps, compressors, production lines, energy systems and material flows. Enterprise AI, by contrast, mostly handles digital workflows such as drafting documents or summarizing meetings.

Global – October 27, 2026 -- A historian export of tags and timestamps is rarely a training-ready dataset for Industrial AI, according to TDengine guidance that tells engineers to establish what each measurement means, when it happened, which asset produced it and whether it is reliable before data reaches an external model-training workflow.

Engineers define the pump decision before selecting data

For a centrifugal pump, the guiding question is whether a reliability engineer should review the pump given its recent operating signals. That differs from asking a model to declare the cause of a failure. Records created after a maintenance decision must not become inputs, because they would not exist when the model makes its prediction.

Global – October 27, 2026 -- A successful OPC UA connection check shows only that a client can reach the endpoint. It does not prove the intended nodes are readable, monitored, mapped or written to the historian, according to a five-step troubleshooting workflow for industrial data pipelines.

The method starts with one known, changing point, traced from the OPC UA server to one expected historian table. Large tag lists hide the first failure.

Global – October 27, 2026 -- Industrial data architects should choose OPC UA when a collector must discover and read a structured source, and MQTT when producers publish through a broker to one or more consumers, according to a TDengine technical comparison. The guide adds that the two names do not describe equivalent layers, so the rule is only a starting point.

OPC UA preserves equipment context that MQTT leaves to the architect

OPC UA is a family of specifications covering Client/Server services, information modeling, Alarms and Conditions, Historical Access and PubSub. A server can organize measurements under equipment objects, attach engineering units and ranges, and expose status, timestamps and methods. A client can browse that address space instead of encoding meaning in topic strings.

Not every server delivers a rich model, however. Some expose flat tag lists, incomplete metadata or vendor-specific structures.

Global – October 27, 2026 -- A pump pressure trend shows when a value left its normal range but cannot show which alarm changed state, what severity the server reported, or when an operator acknowledged it. Those answers sit in separate OPC UA records, and historian designs must decide which questions each record has to answer.

Tag history stores values, not alarm lifecycles

An OPC UA client collects a variable through a monitored item. When DataChange criteria are met, the server sends a DataValue that the historian can store with source timestamp, server timestamp, and StatusCode, depending on the collector mapping. Sampling, deadbands, queues, and timestamp selection determine how much process behavior survives.

Global – October 27, 2026 -- The right OPC UA collection rate is the slowest configuration that still preserves the process behavior needed for operations and historical analysis, and it is set across five separate stages rather than one "scan rate." Device updates, server sampling, subscription publishing, filtering and historian storage each have their own timing. Treating them as one setting can add load without adding information, or hide a pressure transient the historian was expected to explain.

Server sampling cannot beat the source update rate

A PLC may update a temperature every 500 milliseconds while the OPC UA server refreshes its internal value once per second. A client request to sample every 100 milliseconds then re-reads the same value or detects changes no sooner than the server exposes them.