Market and industryNovember 20, 2026Редакция 105.kz

AVEVA linked design and operations: what a single data foundation gives industrial AI

At the AVEVA summit in Taipei, the discussion focused on moving from isolated pilots to scaled industrial AI. The precondition is an agreed data context from SCADA, historians, MES and engineering systems.

Oil & GasEnergyManufacturingPredictive MaintenanceDigital Twins

AVEVA linked design and operations: what a single data foundation gives industrial AI

Key takeaway. The main conclusion of the AVEVA summit: the obstacle to industrial AI is not willingness but infrastructure. Plant data are scattered across HMI, SCADA, MES, historians and engineering systems, and without linking tags to assets, procedures and work orders, models do not produce reliable decisions. For enterprises in Kazakhstan this means starting not with algorithms but with an inventory of sources and an asset register inside the plant perimeter.

What happened

On 29 October 2026, Taipei hosted the 2026 AVEVA Taiwan Technology Summit under the theme "Accelerate Industrial Intelligence, Shape the Future of Industry". As stated in the source material, the company presented industrial AI, the AVEVA CONNECT industrial data platform, digital twins, smart manufacturing and the digitalisation of engineering. The same material cites two assessments of market maturity: according to the "2026 Taiwan Industry AI Transformation Survey" (Artificial Intelligence Technology Foundation), 38.5% of traditional manufacturing companies are at the Ready AI or Scaling AI stages, but only 15.4% have reached scaled adoption; in a survey by the Chung-Hua Institution for Economic Research with a broader definition, 69.0% of manufacturers have already implemented AI, are evaluating implementation or plan to do so. Among those that have implemented it, process automation accounts for 58.6%, the use of IoT and sensor data for 27.8%, and smart manufacturing and lights-out factories for 18.3%.

The central thesis was formulated by Harpreet Gulati, Senior Vice President and Global Head of the PI System business at AVEVA: "For industrial companies, the challenge is not willingness—it's infrastructure". That is, the constraint is not willingness but infrastructure: operational data, engineering data, workflows and compliance requirements are rarely integrated systematically. In support, the material cites an AVEVA and IMD Business School report (2026) based on the views of more than 275 executives from 12 industries: 74% consider the digital ecosystem an important strategic topic, but only 27% take part in substantial data exchange with partners. Product announcements: the Flows module within CONNECT uses more than 800 connectors and processes data in real time at the edge, on-premise or in the cloud; in Q1 2027 the company plans to add an industrial knowledge graph and tools for building digital twins with agentic AI; AVEVA Operations Control allows algorithms written in C# or Python to be deployed at the edge and edge, on-premise and cloud to be combined with data sovereignty in mind.

What this means for enterprises in Kazakhstan

The source lists directly where plant data usually "live": HMI, SCADA, MES, historians and engineering systems — and each is supported by a separate team. Without an agreed industrial context, even data loaded into a model do not produce reliable decisions — this is AVEVA's thesis, and for Kazakhstani enterprises it is recognisable (assessment).

Oil and gas. The practical layer is vibration and temperature of pump and compressor units, process parameters from historians, datasheets and engineering documentation. The task is not to collect signals but to link a tag to a specific asset, procedure and maintenance record; otherwise the model's conclusions cannot be justified at a failure review. See oil and gas solutions.

Power generation. The main body of data is SCADA/EMS, telemetry from feeders and substation equipment, switching logs and ПСВ. Here what matters is not only the accuracy of the RUL estimate but also consistency of data between the dispatch service and the repair service: a discrepancy in the asset register devalues the health index calculation.

Plants. MES, HMI and historian data plus the repair history and operating hours. A typical problem is different identifiers for the same unit in the process control system (АСУ ТП), the maintenance and repair system (ТОиР) and engineering documentation; without bringing them into a single register, predictive analytics remains a set of charts.

Separately on the local context: on-premise and hybrid architectures for national companies remain the norm because of personal data localisation requirements and internal information security policies (common practice), and solutions must work with Russian-language and Kazakh-language documentation and must not require replacing legacy process control systems. The source claims such deployment flexibility but does not measure it.

How to apply: an 8–12 week pilot

  1. Inventory of sources. Compile a list of systems (HMI, SCADA, MES, historian, maintenance and repair, engineering archive): sampling rate, archive depth, owner, tag quality. The output of this step is a data map with named owners, not a model.
  2. Choose one asset and one metric. One unit or one process stage for which failure history already exists and the cost of downtime is clear. The metric is agreed before work starts: unplanned shutdowns, operating hours between repairs, technical availability (КТГ).
  3. Link the context. Map tags to the asset register, link them to procedures and work orders, define data quality rules and a procedure for handling gaps. This is the most labour-intensive part and cannot be deferred to the modelling stage.
  4. Validate on the historical archive. First a retrospective check on accumulated data, then an advisory mode with no automatic actions on the process — with the limits of applicability stated explicitly.
  5. Acceptance and effect measurement. Comparison with a control period against a pre-agreed metric, documented limitations and a decision on scaling. The approach to such work is described in predictive maintenance and digital twins.

It is worth agreeing upfront what counts as the result of a pilot: not a dashboard demonstration but a reproducible assessment of asset condition embedded in a specific workflow — a maintenance request, an inspection round, a repair plan.

Limitations and risks

The figures 38.5%, 15.4%, 69.0% and 74%/27% come from surveys in Taiwan and from an international sample; transferring them to Kazakhstan is not proven — they are an external reference point, not a forecast (assessment). The source is material from the vendor company with a reference to its own report and industry surveys: it records intentions, capability statements and product plans, but does not provide measured effects on reliability, energy consumption or productivity. The announced knowledge graph and digital twin creation tools are stated for Q1 2027, i.e. they are a plan, not a function in operation.

The material contains little data for a full assessment of adoption: there is no information on the quality of source archives, the share of manual entry, the state of engineering documentation or the cost of integration. The practical risks are known: incomplete historians, outdated P&IDs, different registers in the process control system and the maintenance system, restricted access rights between departments. Separately, there is an organisational barrier that the source itself indirectly confirms: announcing data integration is easier than agreeing on an access policy and a data owner within the enterprise.

None of these steps promises a guaranteed economic effect: an 8–12 week pilot tests a hypothesis on your asset and your data, and the decision to roll out is taken on a measured result, not on a demonstration.

Sources

  1. Industrial AI Moves Toward Scaled Deployment: AVEVA Connects Engineering and Operations with a Strong Data Foundation — Kosmo Online · 2026-11-19T19:38:00.000Z
    For industrial companies, the challenge is not willingness—it's infrastructure. To deploy AI safely and at scale, the operational data, engineering data, workflows, and compliance frameworks required are rarely integrated systematically.
  2. Фото: Linda Spashett Storye_book — Bradford Industrial Museum 030.jpg (CC BY 3.0, Wikimedia Commons)

This material was prepared with AI from the sources listed above. Spotted an inaccuracy? Write to info@105.kz.

FAQ

Frequently asked questions

Short, straight answers to the questions we hear most often.

We already have SCADA and a historian. What is missing to start industrial AI?

Following the logic of the source — industrial context: linking tags to specific assets, procedures, work orders and engineering documentation. Without it, even high-quality archives do not allow a model's conclusion to be justified at a failure review.

Can we do without the cloud?

Yes. The AVEVA material states directly that deployment is combined across edge, on-premise and cloud with account for real-time requirements, data sovereignty and compliance. For national companies in Kazakhstan, on-premise remains the baseline option.

How long before a measurable effect?

The source provides no measured effect metrics — only statements about the approach. A realistic format is an 8–12 week pilot with a metric agreed before the start and acceptance against a control period; such a pilot does not guarantee an effect.

What does the announced knowledge graph in CONNECT provide?

It is announced as a tool for mapping existing sources to standard models while preserving data provenance and governance information. It is a Q1 2027 plan, not a feature available today.

Next step

Discuss a pilot project

Tell us about your task — we'll come back with a data audit plan and an effect estimate within two business days.

  1. 1A 30-minute call: the task, your current systems, who makes the decision
  2. 2A review of data and processes, a staged plan with time and cost ranges
  3. 3A 2–4 week prototype or an 8–12 week pilot with success criteria agreed upfront

NDA from the first contact

Reply within one business day