Open automation has been a major focus of the ARC Advisory Group in the US over the past decade. As industrial AI implementation seeks breakthroughs, open automation is experiencing its golden era. The driving force behind open automation, the ARC forum, held a series of global roadshows from February to July 2026 in Orlando (USA), Sitges (Spain), and Bangalore (India), with the Orlando summit serving as the flagship event. The core theme of this year's forum focused on "Scaling Industrial AI Implementation," positioning open automation as the critical underlying technological cornerstone supporting its deployment.
Centered around the two major technological directions of AI applications and open automation, and based on the differentiated characteristics of the manufacturing industry, the forum formed targeted implementation recommendations and industry consensus for the two major fields of discrete manufacturing and process industries. The core takeaways are summarized as follows:
(1) Consensus on Implementation Stages: AI technologies in both industries have officially transitioned from the laboratory pilot validation stage to the production-level scaled application stage. Among them, the progress of scaled implementation in the process industry significantly leads that in discrete manufacturing. The fundamental reason for this difference lies in the varying levels of underlying data maturity between the two types of industries.
(2) Technical Support Logic: Open automation is the underlying architecture supporting the scaled implementation of industrial AI. Whether in discrete manufacturing or the process industry, it is necessary to rely on an industrial data layer built on open standards (such as the ISA-95 enterprise-control system integration standard and the OPC UA communication protocol) to decouple data from hardware assets, thereby ensuring the accurate and reliable operation of AI models.
(3) Industry Scenario Differences: The integration path of AI and open automation must be fully adapted to the differences in key production characteristics between the two industries. The process industry focuses on continuous process optimization and asset reliability improvement as its core objectives; discrete manufacturing, on the other hand, focuses on flexible scheduling in high-mix low-volume scenarios and full-process supply chain collaboration.
(4) Consensus on Core Foundations: Semantic standardization and high-quality governance of industry data are prerequisite conditions for the integrated implementation of AI and open automation. Without this foundation, even the most advanced AI algorithms cannot generate reliable business value.
Another notable feature of this Orlando summit forum is that it established clear industry boundaries for the technical definition of industrial AI for the first time. All implementation cases shared by leading industry enterprises at the forum are distinct from general consumer AI. The core objective of technical implementation is to serve the exclusive constraints of industrial production scenarios, including production process stability, product quality consistency, personnel operation safety assurance, equipment operation reliability, and production environment compliance, rather than merely pursuing the improvement of technical indicators. This definition also marks the formal transition of industrial AI from the concept hype period to the mature period of industry-specific implementation.
Unified Definition of Industrial AI Empowerment
According to the unified definition by the ARC forum, industrial AI technology is not a single algorithm tool, but a comprehensive technology complex covering the entire industrial production process. It includes multiple technologies such as machine learning, deep learning, reinforcement learning, digital twins, computer vision, speech/image recognition, and natural language processing. The deep integration of these technologies with industry process knowledge and production scenarios enables the perception, analysis, prediction, decision-making, and closed-loop optimization of the production process, ultimately forming intelligent production capabilities of self-perception, self-organization, and self-optimization.
The value implementation of industrial AI is absolutely not about forcibly grafting general AI technologies onto production scenarios, but rather requires the technical architecture to be fully adapted to the unique stability, reliability, and real-time constraints of industrial production. Industrial AI is not merely about improving technical indicators, but covering the full-chain scenarios of manufacturing enterprises from underlying production equipment to upper-level business management. The key value directions can be summarized into four major categories:
(1) Asset Performance Management (APM): By analyzing the real-time operation data of equipment to identify fault risks in advance, the traditional passive equipment maintenance mode is transformed into predictive maintenance and condition-based maintenance (CBM) modes, thereby extending the overall effective operation and maintenance time of the equipment and enhancing the value of assets throughout their life cycle. This type of scenario is also the direction with the most significant implementation effects in current industrial AI applications.
(2) Production Process Optimization: Through multi-dimensional joint analysis of real-time process data and historical production data, process parameters are adjusted in real-time and optimized precisely to improve the consistency of product quality, reduce energy and material consumption in the production process, and achieve cost reduction and efficiency increase at the production end. This type of scenario has the most significant implementation effects in the process industry.
(3) Quality Inspection and Control: Through technologies such as computer vision and multivariate statistical process control, real-time quality defect detection is conducted during the production process to promptly identify quality deviations during the production of work-in-progress, completely blocking downstream invalid processing flows. By analyzing the correlation between full-process process parameters and quality results, the process causes of quality fluctuations are precisely located.
(4) Flexible Production and Collaborative Scheduling: Through AI's autonomous perception, decision-making, and execution capabilities, it responds to the production scenario demands of high-mix low-volume, automatically adjusting the process parameters, logistics distribution routes, and production scheduling plans of the production line to achieve rapid product changeover and flexible production adaptation for the entire production line. This type of scenario is the most core AI implementation direction in current discrete manufacturing.
From the perspective of technological evolution trends, the forum also put forward a clear technical judgment: the focus of industrial AI technology applications has rapidly shifted from early perception-based auxiliary applications to closed-loop autonomous decision-making applications based on real-time environmental perception. This also means that the requirements for the completeness, accuracy, and standardization level of underlying data are becoming increasingly stringent.
New Definition and Implementation of Open Automation
In this year's forum, a definition distinct from the past was further clarified: "Based on a vendor-neutral standardized technical architecture system, by decoupling on-site underlying hardware resources from upper-level software application capabilities, production equipment, software applications, and communication networks from different vendors can achieve efficient collaboration and seamless interaction." This is a technological path that subverts traditional automation and breaks the dilemma of traditional automation silos and vendor technology lock-in. After ten years of exploration, the following consensus has been formed: Open automation is not a new proprietary technical architecture dominated by a single vendor, but a collection of "standards of standards" based on existing mature industrial standards; its core logic is not to replace existing industrial standards, but to build on existing standard systems such as OPC UA, ISA-95, ISA-88, and O-PAS to achieve compatibility and collaboration among different standards, ultimately forming a stable technical architecture with clear semantic standards to support the secure and efficient interaction of full-process data.
The core standards and technical frameworks in the current open automation field mainly include three types of directions:
(1) O-PAS Standard Family: Developed and maintained by the Open Process Automation Forum (OPAF) under the Open Group, it is the most core implementation standard framework in the open automation field and also a "standard collection" integrating multiple existing industrial standards. This standard integrates the technical points of multiple existing industrial standards such as ISA-95, ISA-88, OPC UA, IEC 61131-3, and IEC 61499, rather than completely replacing existing standards, achieving seamless interaction among different standards through unified semantic specifications. Its technical architecture has been validated through pilot tests by leading industry users such as ExxonMobil and Lockheed Martin, and has entered the commercial deployment stage.
(2) OPC UA Standard: OPC UA (Open Platform Communications Unified Architecture) has become the de facto standard for secure, platform-independent data exchange in the industrial automation field. It is the most fundamental communication standard in the open automation architecture system, providing a secure, reliable, and standardized industrial data transmission mechanism. It can clearly define the semantic format of data, ensuring standardized data interaction between equipment from different vendors. It is the foundation for achieving efficient data connectivity from underlying field devices to upper-level edge and enterprise cloud platforms, and also the technical prerequisite for achieving IT and OT integration. Over the years, it has successfully achieved secure data interaction among machines, equipment, and software systems in numerous industrial fields such as manufacturing, process automation, energy, and infrastructure.
(3) Joint Support from Industry Standard Organizations: In addition to the above standards, the implementation architecture of open automation also requires supporting from different technical dimensions by industry standard organizations such as the OPC Foundation, the Open Group, UniversalAutomation.org, VDMA, and NEMA. For example, shared runtime technologies, industry standard information models, and industry supporting test and verification standards are provided to ensure interoperability in multi-vendor environments for the open automation architecture.
At the same time, industry users and suppliers have also generally realized that open automation is not merely a simple iteration of technical architecture, but a reconstruction of the technological ecology of the entire industrial automation industry—redefining the traditional vendor lock-in dependency model into a user-centric optimal technology combination model based on standards. This transformation will completely release the value space for industrial AI implementation. The industrial automation industry is increasingly aware that this technological transformation is not a single-point technical breakthrough, but a systematic reconstruction involving technical architecture, industry ecology, project implementation, and business models. The core characteristics of the open automation system can be summarized in three dimensions: first, vendor neutrality, achieving seamless interchange, integration, and collaboration among equipment from different vendors through unified open standard interfaces; second, software definition, completely decoupling control layer software from underlying hardware resources so that software is no longer bound to proprietary hardware platforms of specific vendors; third, security by design, strictly designing the security protection system in accordance with international industrial cybersecurity standards such as IEC 62443, embedding risk protection mechanisms into all hierarchical levels of the architecture, rather than overlaying security protection solutions on existing architectures.
From the perspective of industry validation progress, there are obvious differences in the degree of technology acceptance and implementation depth among different industries. This difference is directly related to the industry's tolerance for technical barriers, security protection, and operation and maintenance costs. Among them, the petrochemical industry, as a typical field with the most prominent pain points of traditional closed systems, is currently the most mature field for technology implementation. This is because the petrochemical industry is a typical capital-intensive process industry. Its production scenarios have core characteristics such as high temperature and high pressure, flammable and explosive materials, and continuous and uninterrupted production processes, which place extremely high requirements on the real-time control accuracy, operation stability, and security protection capabilities of automation systems. Moreover, the life cycle of production facilities in the petrochemical industry is generally as long as 30 to 50 years, with frequent demands for subsequent technological upgrades, transformation, and expansion. This industry characteristic precisely hits the important technical pain points of traditional closed automation systems. In discrete manufacturing, the open architecture of industrial automation has shifted from the concept validation stage to the industry-scale implementation stage. It is decoupling traditional proprietary automation control capabilities through software, connecting the data links between multi-source equipment and high-level systems through standard protocols, and ultimately supporting discrete manufacturing enterprises in transforming the four core values of flexible production, rapid changeover, global equipment interconnection, and data-driven optimization from engineering concepts into quantifiable production results. The food and beverage industry and the pharmaceutical industry have strict requirements for production compliance and data traceability, and have currently also entered the stage of substantive deployment. In contrast, there are relatively few public validation cases in the power industry, which is still in the stage of technical solution demonstration and local scenario pilot testing.
From the perspective of industry implementation progress, the open process automation architecture of O-PAS has entered the industry-level scaled validation stage. The world's first commercial-grade lighthouse project, led by ExxonMobil and with Lockheed Martin participating in system integration, has been successfully and stably running at ExxonMobil's resin refinery. As of July 2026, the system has been running stably for more than 18 consecutive months without any unplanned downtime caused by system architecture or multi-vendor equipment adaptation issues. Feedback data from on-site operation and maintenance personnel shows that the daily operation and maintenance workload of this system has been significantly reduced compared to the original traditional DCS system, which directly verifies the advantages of the open architecture in simplifying operation and maintenance and reducing fault handling time. Heaven Hill Bourbon Distillery in the food and beverage industry and the Spanish pharmaceutical company Rovi in the pharmaceutical industry have also successively completed new system deployments or old system upgrades based on open automation standards. Both projects adopt the PlantPAx 5.1 distributed control system launched by Rockwell Automation for the digital transformation of the process industry. The system architecture concept aligns with the development direction of O-PAS, but it has not yet fully implemented all the requirements of the O-PAS reference architecture. The core process control logic of its platform relies on the Logix controller ecosystem and adopts the IEC 61131-3 programming system. Configuration and control application migration do not rely on the IEC 61499 standard, and control applications cannot be freely migrated out of the Rockwell hardware environment. Third-party hardware can only be accessed as field devices and cannot serve as Distributed Control Nodes (DCN) to carry real-time process control. In addition, the system lacks the vendor-neutral independent orchestration layer defined by O-PAS, and batch scheduling and system collaboration capabilities are bound to the FactoryTalk software suite.
As of July 2026, more than 10 cross-industry open automation benchmark projects have completed deployment or acceptance globally, fully validating the industry adaptability, commercial implementation value, and scalability of this technical architecture.
Value Reconstruction of Open Automation Empowering Industrial AI Implementation
Industrial AI technology is rapidly penetrating from concept validation to scaled implementation, but the actual implementation situation in the industry is lower than expected. Most industrial AI projects are still concentrated in benchmark scenarios of leading enterprises. After completing laboratory validation or pilot scenario validation, many enterprises' industrial AI projects cannot be quickly promoted in plant-wide or multi-plant scenarios. The technical bottleneck leading to this result is the prevalent "data silos" in the industrial industry, that is, the "closedness" problem of automation systems. Under the traditional closed architecture, equipment from different vendors adopts different communication protocols, data formats, and command line logic. Data is locked in various independent subsystems and cannot be uniformly collected and parsed. This means that industrial AI applications struggle to obtain high-quality training data that conforms to unified standards. Without the support of such high-quality data, no matter how advanced the AI model is, it cannot be implemented in actual production scenarios.
However, a core value of open automation technology lies in its ability to connect all data circulation links from the field device layer to the upper business layer through standardized technical interfaces, building an "open, standardized, and unified" data backbone. This capability precisely hits the core pain point of industrial AI implementation and is the key technical support for breaking data silos and achieving large-scale, scaled implementation of industrial AI technology. (See Figure 1)
Figure 1 Dual-Wheel Drive of Open Automation and Industrial AI
From the perspective of actual implementation, the technical value of open automation in industrial AI implementation is mainly concentrated in three aspects:
(1) Breaking Data Silos: Achieving unified data interoperability is the most basic and core value of open automation technology for industrial AI implementation. The training data for industrial AI applications comes from different vendors and different types of equipment at the production site. The communication protocols and data formats of such equipment are often different. If the data from such equipment cannot be collected and parsed according to unified standards, subsequent industrial AI applications will not be able to obtain high-quality training data support, let alone implementation applications. One of the core technologies of open automation is the unified standardized communication framework based on OPC UA. This framework is equivalent to providing all equipment with a unified "data interaction translation mechanism." It can uniformly encapsulate field device data from different vendors, different communication protocols, and different data formats into a format that complies with OPC UA standards without changing the operating state of existing equipment, achieving seamless full-link data connectivity from the field device layer to the upper business layer. This capability precisely solves the core data bottleneck of industrial AI implementation. From actual implementation cases, the supporting effect of this technical value has been fully validated in multiple industry-level projects. The validation results of these actual projects fully prove the value of open automation technology in achieving data interconnection and supporting industrial AI implementation. With the standard data support of the open architecture, the application deployment cycle of industrial AI can be significantly shortened. Without this data foundation, the scaled implementation of industrial AI will be out of the question.
(2) Develop Once, Run Everywhere: The open automation architecture is the engineering support for breaking the bottleneck of industrial AI implementation and achieving scaled reuse and promotion of models. It must first be clarified that the cross-scenario reuse capability of "develop once, run everywhere" is currently most maturely validated mainly in the process industry.
In the process of industrial AI industrialization and implementation, insufficient model portability is a common technical problem in the industry, which greatly restricts AI technology from moving from single-point pilots to full-domain scaled applications. Under the traditional closed architecture system of industrial automation, there are significant differences in the brand and model of automation equipment, underlying communication protocols, data collection formats, and data interface specifications in different manufacturing scenarios and factory systems, forming fragmented technical barriers. This is the underlying root cause leading to the difficulty of cross-scenario reuse of industrial AI. Industrial AI models developed and trained based on closed architectures have strong scenario-specific binding attributes. After the model completes validation and implementation on a single production line or a single factory, if it needs to be reused in other production scenarios in the same industry and with the same process, it often requires comprehensive adaptation and transformation for the equipment system, protocol standards, and data specifications of the target scenario. In most scenarios, it is not only necessary to iteratively adjust the model data access logic and algorithm preprocessing rules, but even to re-conduct on-site data collection, dataset cleaning, model training, and parameter tuning. This adaptation process has high technical complexity, high labor and time costs, and an implementation cycle of up to several months, which greatly reduces the reuse efficiency and industrialization implementation speed of industrial AI projects, and is a key pain point hindering the scaled popularization of industrial AI (see Figure 2).
Relying on the layered open architecture built on the O-PAS open automation standard system, coupled with unified standardized interface capabilities, the development and deployment logic of industrial AI models has been thoroughly reconstructed, truly achieving the portability capability of industrial AI models to develop once, run everywhere, providing an implementable technical foundation for the scaled implementation of industrial AI. However, it must be clarified that the reuse support of this architecture for industrial AI models is currently only applicable to production scenarios with the same process but different control systems within the process industry. The prerequisite for reuse is that the underlying working principles of the process equipment are consistent, allowing for differences in single-plant capacity and design output. The core prerequisite for cross-scenario reuse is data semantic consistency, which is provided by the standard specification layer of the open automation architecture with rigid guarantee. Its technical logic lies in the fact that the open automation architecture can uniformly encapsulate and standardize the underlying communication protocols and private data formats of various heterogeneous automation equipment on site, adapting them globally to a universal and unified OPC UA standard data format, achieving complete decoupling between the AI application layer and the underlying technical details of the equipment. That is, the model does not directly interface with the equipment, but only interfaces with the standard data interface layer of the architecture, and the underlying equipment differences are completely isolated outside the model logic. Based on this architecture, there is no need to adapt to the equipment parameters and private protocols of various scenarios during the development stage of industrial AI models. It only needs to interface with the unified OPC UA standardized data interface and standardized semantic paradigm to complete model training, iteration, and effect validation.
Figure 2 Comparison of Scaled Implementation of Industrial AI under Traditional Closed Architecture and Open Automation Architecture
In the model deployment and reuse stage, as long as the automation system of the target production scenario complies with the O-PAS open architecture standard and belongs to the same process category within the process industry, the trained mature industrial AI model can be directly accessed into the scenario unified orchestration layer. Through the standard data adaptation capability of the architecture, it can read the standardized process data and equipment operation data of the target scenario in real-time, completing preset core functions such as intelligent analysis, process optimization, and reasoning decision-making. The entire process does not require modification and adaptation of the model algorithm logic and reasoning decision rules, nor does it require re-collecting on-site data and training the model. It only needs to complete the basic data mapping configuration on the architecture side to achieve rapid implementation. This technical characteristic greatly shortens the long model cross-scenario adaptation cycle under the traditional closed architecture, greatly reduces the cross-scenario migration cost of industrial AI models, and significantly improves the scaled deployment efficiency of multiple scenarios, multiple factories, and multiple bases in the process industry.
From the perspective of industry technical validation achievements, the global open automation ecosystem has fully validated the implementation feasibility and technical value of this architecture. Relying on the open automation system built on the O-PAS standard, leading industry automation enterprises, together with core end-users in the industrial chain, have carried out multiple rounds of open architecture adaptation and AI model migration validation. Practice has proved that industrial AI models for process optimization, equipment predictive maintenance, and production quality inspection developed based on standardized open architectures can effectively shield the technical differences of automation equipment from different vendors and differentiated on-site operating environments, achieving cross-scenario non-differentiated reuse in scenarios with the same process but different control systems within the process industry. Moreover, the process optimization amplitude, decision response speed, and operation and maintenance support effects after model implementation remain stable and consistent with the original validation scenarios, fully corroborating that the open automation architecture is the core technical support for industrial AI to break through scenario barriers and achieve scaled industrialization implementation in the process industry field.
(3) Bridging the Last Mile to Support Closed-Loop Real-Time Decision-Making: This is the core business value of open automation technology for industrial AI implementation. The business value of industrial AI applications is reflected in "closed-loop real-time decision-making," that is, the AI system conducts real-time analysis and reasoning decision-making based on the equipment operation data and process parameters collected in real-time on site, and issues the decision results to the on-site automation control equipment in real-time to complete automatic process adjustments. This closed-loop process of "data collection-analysis and decision-making-execution and adjustment" must be completed in an extremely short time. Otherwise, real-time state changes in the production scenario will lead to the invalidation of decision results, or even trigger safety accidents. Under the traditional closed architecture, this technical requirement is difficult to meet. Due to the lack of unified equipment communication protocols and data formats, the control instructions issued by the upper business system need to undergo multi-layer protocol conversion before reaching the field equipment. The delay time of this process cannot meet the requirements of real-time control. In many cases, the decision results of industrial AI can only be presented to operation and maintenance personnel in the form of "reference suggestions" and manually executed by them, without truly closing the loop between "data decision-making" and "actual execution." However, open automation technology provides core support for closed-loop real-time decision-making for industrial AI applications through the full-link standard OPC UA communication architecture and the localized computing support of the distributed control layer. This capability is a key prerequisite for industrial AI to realize business value. (See Figure 3)
Figure 3 Closed-Loop Empowerment System from Data Foundation to Intelligent Decision Support
Specifically, the technical logic of the open architecture supporting closed-loop real-time decision-making is divided into two levels: low-latency support for full-link standard communication and local computing support for the distributed control layer. Under the open architecture, all communications from the field device layer to the upper orchestration layer adopt the standard OPC UA protocol. Based on extended specifications such as OPC UA FX, coupled with local computing of DCN distributed control nodes, the end-to-end delay of control instructions can be compressed to the millisecond level, meeting the real-time requirements of industrial closed-loop control. Under the open architecture, the distributed control node DCN has edge computing capabilities and can conduct real-time analysis of the collected process data at the production site. The reasoning and decision-making tasks are sunk to the edge side close to the field equipment for completion. Only complex process optimization tasks that require long-term data support will be uploaded to the upper cloud business system for completion. This hierarchical computing architecture further compresses the delay of data transmission, allowing decision instructions to be issued to field equipment faster to complete real-time adjustments. This technical value has been fully validated in multiple industry-level projects.
Integration and Symbiosis
The integration and symbiosis logic between open automation and industrial AI needs to be clearly defined. Open automation is the foundational support for the scaled implementation of industrial AI. Industrial AI is the core outlet for improving high-level business value after realizing the open automation architecture. The two must be implemented in combination and cannot realize value transformation independently (see Figure 4). This integration logic can also be summarized as: industrial AI without open automation is difficult to form a system, and open automation without AI lacks incremental value. Of course, this does not exclude a large number of successful applications where AI acts on non-standard data (such as vibration waveforms, acoustic fingerprints, infrared thermal images, etc.). The data governance standardization provided by open automation is a sufficient condition for the success of industrial AI, not a necessary condition. In short, open automation is the optimal engineering path for industrial AI to achieve replicable and scalable implementation.
Specifically, the integration logic of the two is reflected in two dimensions:
(1) Open automation is the foundation for scaled AI implementation: Open automation unifies multi-source heterogeneous data among underlying equipment into a standard data format with contextual business semantics through standardized semantic models and communication protocols. In particular, the specifications of the ISA-95 standard hierarchical architecture and information flow, as well as the semantic standardization carried by the ISA-88 (batch processing) or MES/MOM ontology model, along with the supporting information model of OPC UA, provide high-quality and highly consistent standardized data support for AI models, which is the basic condition for AI technology implementation. Without the standardized data brought by open automation, AI models must be trained and inferred based on non-standard heterogeneous data, which has a very high probability of outputting unreliable results.
(2) AI is the core value scenario for open automation: The standardized data brought by open automation can only be transformed into actual business value of improving production efficiency, enhancing product quality, and reducing production costs through the analysis, optimization, and decision-making capabilities of AI technology. Without the value transformation support of AI technology, the open automation architecture itself will only produce basic system integration efficiency improvements and cannot release greater business value.
Figure 4 Technological Combination of Mutual Support between Open Automation and Industrial AI
During the 2026 ARC Industry Forum, after in-depth discussions involving all relevant parties including end-users, technology suppliers, and industry standard organizations, a complete implementation consensus on AI empowerment in the process industry and discrete manufacturing industry was formed: Open automation is a prerequisite for the scaled implementation of AI. A solid open data foundation must be prioritized so that AI technology can deliver quantifiable and scalable business value.
Its core logic is that the biggest difference between AI technology implementation in industrial scenarios and Internet-type AI technology is that industrial scenarios have extreme requirements for the reliability of results, and no deviations can occur. If the data supporting the AI model is one-sided, intermittent, non-uniform in format, and lacks semantic association, no matter how advanced the AI model is, it cannot output reliable results that meet industrial production constraints, and may even produce serious negative effects. It is reported that among enterprises that have already scaled AI technology implementation, over 80% of them first built an industrial data governance system based on open automation standards before implementing AI technology. Among enterprises still in the AI pilot stage, over 70% focus their energy on optimizing model algorithms rather than data foundation governance, which is also the core reason why their pilots cannot be scaled and replicated. This fact reveals that a key value of the open automation architecture is to provide a standard data governance system that unifies multi-source heterogeneous data from different equipment and different systems into standard data with complete business semantics. Without the support of this foundational data, the implementation results of industrial AI technology cannot meet industrial-grade reliability requirements.
The integration path of AI and open automation must be adapted to the production characteristics of the industry. There is no universal "all-purpose" integration solution. The combined implementation of the two technologies must anchor the actual business pain points of different industries, fully adapt to the special constraints of production scenarios, and the technical solution should be defined by business pain points rather than by the products of technology suppliers. According to the implementation scenario characteristics of the two types of industries, process and discrete manufacturing, it is clear that there are significant differences in the comparison of technical priorities between the two types of industries, and differentiated implementation must be adapted to industry scenarios (see Table 1 for details).
Table 1: Implementation Differentiation Adapted to Discrete Manufacturing and Process Industry Scenarios
| Dimension | Discrete Manufacturing | Process Industry |
| Core Industry Scenario Objectives | Flexible production adaptation, rapid production line changeover, supply chain collaborative scheduling, quality process control, asset operation and maintenance optimization | Real-time optimization of process parameters, full-process multivariate constraint control, asset operation and maintenance optimization, continuous quality control, energy/material consumption optimization |
| Open Automation Implementation Priority | High: Need to decouple hardware and upper-level software applications to support flexible production resource scheduling | Extremely High: Need to integrate legacy systems and build standardized redundant data collection channels to support full-link monitoring of continuous process processes |
| AI Application Implementation Priority | Medium: Relying on standard data provided by open automation, basic architecture adaptation must be completed first before implementing AI applications | High: The value return of process optimization directly depends on data quality. The open automation standard architecture must be implemented first before deploying AI applications on it |
| Core Dependencies of AI Applications | Production process parameters, position location data, visual inspection data, equipment operation status data | Process parameter analysis, process mechanism modeling, association of equipment operation status data with process context |
| Difficulties in Integrated Implementation | Standard integration of multi-source heterogeneous data, standardized adaptation of process equipment from different vendors, collaborative docking of a large number of IT systems | High real-time collection of continuous process data, standard integration of multi-source heterogeneous data, complete time-series association of long-cycle data, dynamic coupling of multiple process parameters |
The implementation of the open automation architecture must rely on standards and vendor-agnostic adaptation, including standard adaptation, vendor neutrality, and gradual transformation of existing legacy systems.
Standard Adaptation: The standard combination that the open automation architecture must adapt to includes: OPC UA is the core unified communication standard, ISA-95 is the core standard for industrial data hierarchical modeling and establishing data semantic and contextual consistency, and the O-PAS standard is the basis for the overall architecture implementation in the process industry. The semantic definitions of all standards must be unified within the enterprise, and there must be no inconsistencies in the semantic definitions of the same data by different systems.
Vendor Neutrality: The implementation of the open automation architecture must fully comply with the core principle of "vendor neutrality." In the architecture design, there must be no binding dependency on the technology or products of any specific vendor. Instead, seamless collaboration among equipment from different vendors must be achieved based on industry standard interfaces, leaving sufficient flexibility for subsequent system upgrades and expansions.
Gradual Transformation of Existing Legacy Systems: Enterprises should not adopt a "tear down and start over" approach to directly replace existing legacy systems that are still running stably. Instead, under the premise of "not affecting the operation of existing systems," they should use standard interface servers, edge collection units, and other equipment to access the original system data into the new architecture, achieving long-term collaborative coexistence of old and new systems. For subsequently newly launched control systems, equipment supporting open standards must be prioritized to gradually complete the standardized transformation of the entire plant architecture.
Consensus on Application Implementation Sequence: From Basic to High-Level Value
Based on the implementation practice experience accumulated by some leading enterprises, a priority sequence for the scaled implementation of AI technology has gradually been formed. That is, AI implementation needs to follow an improvement path from basic to complex and from auxiliary to core, advancing in stages and step by step. It cannot directly start with high-level value scenarios such as core process optimization (see Figure 5).
Figure 5 Industrial AI Implementation Path: From Basic to High-Level Value
The specific implementation sequence is divided into four levels (see Figure 6):
L1 (Basic Visualization): First, through the open automation architecture, complete the standardized collection and integration of underlying equipment data, and build a complete industrial data visual monitoring platform. This allows production and operation and maintenance personnel to obtain real-time perceptual information such as the operation status, process parameters, and production environment of on-site equipment through the monitoring platform, achieving transparent management of the production site. This is the basic prerequisite for all AI implementations.
L2 (Auxiliary Analysis): Based on standardized historical data, conduct descriptive analysis, mine historical data through AI models, locate the root causes of various problems in the production process, and find key directions for optimization. Present the analysis results to production and operation and maintenance personnel in a visual way to assist them in production decision-making. The implementation risk at this stage is extremely low, the value can be quantified, and it can also verify the completeness and standardization of the data foundation.
L3 (Predictive Maintenance): After the maturity of the data foundation is further improved, AI technology is further extended from analytical applications to predictive maintenance scenarios for critical assets. Through real-time analysis of equipment operation data by AI models, hidden fault risks of equipment are identified in advance. This is currently the most maturely validated and clearly value-producing AI implementation scenario in the industry.
L4 (Closed-Loop Autonomous Decision-Making): After completing the scenario validation of the first three stages and when the maturity of the data foundation is completely stable, finally implement the most complex closed-loop autonomous decision-making scenarios such as process parameter optimization and global resource scheduling. The AI model transforms analysis results into adjustment decisions and directly issues them to the underlying control systems, achieving real-time adjustment of production processes and dynamic scheduling of production resources. This type of scenario has the greatest business value and the highest implementation difficulty, requiring complete standard data foundation support.
Figure 6 Four Stages of Scaled Implementation of Industrial AI
This phased implementation sequence is a summary of successful practices by leading industry enterprises. Enterprises must follow a step-by-step path from basic to high-level value for the scaled implementation of AI technology to achieve actual results.
Concluding Remarks
The development stage of industrial AI technology has completed a qualitative change. AI technologies in both the process industry and discrete manufacturing have substantively moved from the laboratory pilot exploration stage to the scaled production-level implementation stage. However, the core constraint for value release is whether there is support from complete, accurate, semantically standardized, and high-quality governed industry data. This is the most critical prerequisite for determining the effectiveness of AI implementation.
And open automation is the priority engineering path for industrial AI empowerment. To support the industrial scaled implementation of AI technology, the open automation architecture can thoroughly break the information silos of traditional automation systems and achieve standardized data collection, integration, and governance. Without this foundational support, the implementation effects of industrial AI will be greatly compressed, or even directly fail.
In this process of technological integration, it is essential to adapt to industry characteristics. There is no universal AI and open automation integration solution. The implementation strategy must anchor the core production pain points of the industry and fully adapt to the special constraints of industry scenarios. For the process industry, its continuous production characteristics determine that the implementation priority of open automation is higher. The process data of existing legacy systems must first be integrated through the O-PAS standard architecture, and then AI applications such as process optimization, predictive maintenance, and full-process multivariate constraint control are deployed based on the standard data. For discrete manufacturing, its pain points of high-mix and low-volume production determine that flexible production scenarios should be the core main line, promoting the implementation of open automation and AI step by step. Pilot testing should first be conducted on core production lines, achieving data standardization through the OPC UA standard architecture, and then gradually expanding to the entire plant. AI technology should be prioritized for implementation in scenarios such as visual inspection, asset operation and maintenance, and supply chain scheduling, and then extended to flexible production changeover and process parameter optimization scenarios.
(End of the article)
About the Author
Peng Yu: Shanghai Institute of Process Automation Instrumentation, PLCopen China Organization