MINI REVIEW article

Front. Mech. Eng., 23 July 2026

Sec. Digital Manufacturing

Volume 12 - 2026 | https://doi.org/10.3389/fmech.2026.1842661

A review of intelligent decision-making and planning methods for machining process routes

  • 1. School of Mechanical Engineering, Shandong Huayu University of Technology, Dezhou, China

  • 2. Department of Mechanical and Manufacturing Engineering, Faculty of Engineering and Built Environment, Universiti Kebangsaan Malaysia, Bangi, Selangor, Malaysia

  • 3. Shanghai Marine Equipment Research Institute, Shanghai, China

Abstract

Intelligent decision-making and planning for machining process routes play a critical role in bridging product design and manufacturing, with direct implications for machining efficiency, production cost, and product quality. However, traditional process planning relies heavily on manual expertise and is increasingly unable to meet the demands of multi-variety, small-batch, and customized production in intelligent manufacturing. This review examines the main methodological paradigms in intelligent machining process planning, including knowledge-driven, algorithm-optimization-based, data-driven, and hybrid approaches, and discusses their principles, strengths, limitations, and application scenarios. It also reviews the roles of enabling technologies such as model-based definition, knowledge graphs, and digital twins in supporting process knowledge organization, route generation, and dynamic adaptation. On this basis, the current challenges of intelligent machining process planning are analyzed from the perspectives of knowledge representation, optimization robustness, data quality, system integration, and industrial deployment. Finally, future development trends are outlined toward more knowledge-enhanced, adaptive, integrated, and practically deployable planning frameworks. This review aims to provide a concise reference for future research and engineering application in intelligent machining process planning.

1 Introduction

Manufacturing is being reshaped by Industry 4.0 and the growing demand for personalized, make-to-order production. In this context, machining process planning plays a critical role because it directly affects production efficiency, manufacturing cost, and product quality (; ). However, traditional process planning still relies heavily on human experience and therefore often suffers from subjectivity, low efficiency, and poor consistency. These limitations make it difficult to handle complex parts and dynamic production environments (). Recent advances in artificial intelligence, big data, and related digital technologies provide new opportunities for intelligent process planning (). By integrating multi-source manufacturing knowledge and production data, these technologies can support more efficient and accurate route generation while reducing dependence on manual expertise. A systematic review of current methods, their strengths and limitations, and their application scenarios is therefore necessary for both academic research and engineering practice.

Intelligent machining process planning aims to optimize multiple objectives, such as machining time, cost, energy consumption, and quality, while satisfying process constraints such as datum-first and rough-before-finish principles (). To achieve this goal, several key issues must be addressed, including process knowledge representation, machining feature recognition, constraint handling, and multi-objective decision-making (). Intelligent process planning is a key component of intelligent manufacturing. Existing methods differ in their underlying decision logic, ranging from experience and rule reuse to optimization search, data learning, and multi-paradigm fusion (). Figure 1 summarizes the overall methodological framework of intelligent process planning for machining process routes.

FIGURE 1

This review organizes existing studies into four main categories: knowledge-driven, algorithm-optimization-based, data-driven, and hybrid approaches. It further examines the roles of enabling technologies, including model-based definition (MBD), knowledge graphs, digital twins, reinforcement learning, and large language models, in machining process route planning. The review was based on a targeted and reproducible literature search in Web of Science and Google Scholar. The search strings combined process-planning terms with method and technology terms, including (“machining process planning” OR “computer-aided process planning” OR CAPP) AND (“knowledge-driven” OR “knowledge graph” OR “optimization” OR “data-driven” OR “reinforcement learning” OR “digital twin” OR “large language model”). Recent publications from 2018 to 2026 were prioritized, while earlier foundational studies were retained when they were necessary to clarify the methodological origins of CAPP and intelligent process planning. Records were screened in two stages: first by title and abstract for relevance to machining process planning, and then by full-text assessment for methodological relevance to route generation, process reasoning, optimization, data-driven recommendation, or enabling digital technologies. The final cited corpus contained 47 publications. Inclusion criteria were: direct relevance to machining process route planning or CAPP, contribution to a representative methodological paradigm or enabling technology, and explicit discussion of planning logic, required inputs, constraints, evaluation, or application scenarios. Exclusion criteria were: studies unrelated to machining process planning, purely application-specific reports without transferable methodological value, papers focused only on machining parameter optimization without route-planning implications, and papers lacking sufficient methodological information.

Compared with previous reviews on CAPP and intelligent process planning, this mini-review does not focus only on conventional CAPP architectures, feature recognition, or a single class of optimization algorithms. Instead, it compares four methodological paradigms from the perspectives of decision logic, required input data, suitable applications, limitations, and industrial deployability. It also links emerging technologies such as knowledge graphs, digital twins, reinforcement learning, and large language models specifically to machining process route planning rather than treating them as general artificial-intelligence tools.

2 Major approaches to intelligent process planning

2.1 Conventional knowledge- and optimization-based approaches

2.1.1 Knowledge-driven approaches

Knowledge-driven approaches represent the earliest and most interpretable paradigm in intelligent process planning (; ). Their central idea is to formalize process knowledge from expert experience, handbooks, design rules, and historical cases into computable representations, so that feasible process routes can be generated through reasoning and matching mechanisms. The main advantage of this paradigm lies in its transparency: planning logic can be explicitly traced, and process constraints can be incorporated in a relatively controllable manner ().

Early studies mainly relied on rule-based reasoning, in which if–then rules were constructed to map part features and manufacturing constraints to corresponding machining operations. Such methods are suitable for relatively well-structured parts and stable production scenarios because they are straightforward to implement and easy to interpret (; ). However, their limitations become increasingly evident as part geometry, constraint coupling, and production variability increase. In particular, rule conflicts, rule explosion, and difficult maintenance often reduce their scalability and practical adaptability.

Case-based reasoning was later introduced to alleviate these shortcomings. Instead of depending entirely on manually encoded rules, case-based methods retrieve and adapt process routes from similar historical parts. This strategy is more flexible for customized and complex parts, especially when direct rule formulation is difficult (; ). Nevertheless, its effectiveness strongly depends on the completeness of the case base, the quality of similarity evaluation, and the reliability of case adaptation. As a result, the method is often constrained by data accumulation and knowledge reuse efficiency.

More recently, knowledge graphs have extended knowledge-driven planning from simple rule or case retrieval to structured semantic reasoning. By organizing heterogeneous process knowledge into explicit entity–relation networks, knowledge graphs improve knowledge reuse, semantic association, and process recommendation. This makes them particularly valuable in complex domains where process knowledge is distributed across multiple sources and cannot be efficiently represented in traditional rule sets (). However, large-scale knowledge construction, dynamic updating, and consistency maintenance remain major challenges. Therefore, although knowledge-driven approaches retain clear advantages in interpretability and constraint control, their industrial scalability still depends on more efficient knowledge organization and maintenance mechanisms.

2.1.2 Algorithm-optimization-based approaches

Algorithm-optimization-based approaches formulate process planning as a combinatorial optimization problem under multiple objectives and constraints. Typical optimization targets include machining time, production cost, energy consumption, and quality, while constraints may involve operation precedence, machine availability, datum relationships, and process feasibility (). Compared with knowledge-driven methods, this paradigm places greater emphasis on solution quality under complex objective trade-offs.

Traditional heuristic and metaheuristic algorithms, such as genetic algorithms, ant colony optimization, particle swarm optimization, and simulated annealing, have been widely applied in this context. Their major strength is the ability to search large solution spaces and obtain near-optimal solutions within reasonable computational time. For this reason, they are particularly useful when process planning cannot be expressed by explicit rules alone and must instead be solved through structured search.

However, the performance of individual algorithms is often limited by their intrinsic search characteristics. Some methods exhibit good global search capability but converge slowly, whereas others converge quickly but are more vulnerable to premature convergence or local optima. In multi-objective scenarios, these weaknesses become more pronounced because the algorithm must not only identify feasible routes but also maintain diversity and robustness across competing objectives ().

To overcome these limitations, many studies have adopted hybrid optimization strategies that combine complementary search mechanisms or integrate intelligent optimization with fuzzy reasoning and Pareto-based decision methods (). In addition, reinforcement-learning-based optimization has attracted growing attention because it provides a potential route from static route generation to adaptive online decision-making (). Even so, algorithm-optimization-based approaches still face three major issues: model formulation is often problem-specific, parameter tuning can be difficult, and transferability across different part categories and production contexts remains limited. In other words, these methods are powerful for optimization, but their practical effectiveness depends heavily on how well the planning problem is modeled. A broader cross-paradigm comparison is provided in Table 1.

TABLE 1

ApproachDecision logicRequired input dataSuitable applicationsMain limitationsIndustrial applicability
Knowledge-drivenExplicit rules, cases, ontologies, or knowledge graphs are used for reasoning and route recommendationPart features, MBD/CAD information, process rules, handbooks, historical cases, machine and tool knowledgeStable product families, rule-constrained planning, and tasks requiring explainable decisionsKnowledge acquisition is costly; rule conflicts, rule explosion, and dynamic updating remain difficultHigh interpretability; practical when the knowledge base is maintained as a living enterprise asset
Optimization-basedProcess planning is formulated as a constrained combinatorial or multi-objective search problemCandidate operations, precedence constraints, machines, tools, time/cost/energy models, objective weightsRoute selection under competing objectives such as machining time, cost, energy use, and qualityProblem-specific modeling, parameter tuning, premature convergence, and limited transferabilityUseful for offline decision support; stronger deployment requires integration with real-time shop-floor data
Data-drivenHistorical data are used to learn mappings from part/process features to planning decisionsLabeled process plans, CAD-derived features, quality data, equipment data, and production recordsProcess recommendation, feature recognition, route reuse, and adaptation in variable production contextsPerformance depends on data quality, labeling consistency, interpretability, and validationPromising but still limited by data governance, model trustworthiness, and engineering verification
HybridKnowledge constraints, optimization search, and data learning are combined in one planning architectureStructured knowledge, historical cases, optimization models, production feedback, and digital-twin dataComplex parts, mixed-variety production, dynamic resources, and closed-loop planningSystem complexity, module coordination, interoperability, and maintenance burden are highLikely the most practical route for industrial deployment, but requires standard interfaces and robust validation

Critical comparison of major paradigms for intelligent machining process route planning.

2.2 Emerging data-driven and hybrid approaches

2.2.1 Data-driven approaches

Data-driven approaches have become one of the most active research directions in intelligent process planning. Instead of relying primarily on explicit expert knowledge, they use manufacturing data from design, process execution, equipment operation, and quality inspection to support route generation, decision-making, and parameter optimization (; ). This shift reflects a broader transition from experience-dependent planning toward learning-based planning.

In machining process route planning, machine-learning-based methods are mainly used to learn mappings from part features, materials, tolerances, historical process plans, and shop-floor records to candidate operations, machines, tools, or process sequences (). Deep-learning-based methods are more relevant when the input information is high-dimensional or weakly structured, such as CAD-derived features, process documents, or graph-structured manufacturing knowledge (). Reinforcement learning is directly linked to process route planning because a route can be formulated as a sequential decision-making problem: each operation selection changes the subsequent feasible actions under precedence, resource, and process-feasibility constraints (). Recent studies have also explored large language models and knowledge-enhanced foundation models for process-knowledge extraction, semantic interaction, process recommendation, and multi-agent route generation (; ). However, in machining process planning, these models should be treated as decision-support components rather than autonomous planners, because generated plans still require grounding by knowledge graphs, process rules, constraint checking, or simulation-based verification to avoid infeasible routes.

Despite these advances, data-driven approaches still face several important limitations. Their performance is strongly influenced by data quality, data volume, and labeling consistency. In many machining scenarios, available datasets remain fragmented, small-scale, or weakly standardized (). Moreover, limited interpretability and uncertain generalization make these models difficult to validate in safety-critical or high-value manufacturing environments. Therefore, while data-driven planning offers strong adaptability and learning capacity, its broader application still requires improvements in trustworthy modeling, data governance, and engineering validation.

2.2.2 Hybrid approaches

Hybrid approaches combine knowledge-driven, optimization-based, and data-driven methods in order to compensate for the weaknesses of individual paradigms. This has become an increasingly important direction because process planning in real manufacturing environments rarely depends on a single decision mechanism (). Instead, it usually requires simultaneous consideration of process rules, optimization objectives, data patterns, and dynamic disturbances.

A typical hybrid strategy is to use prior knowledge to constrain the feasible solution space, optimization algorithms to search for high-quality process routes (), and data-driven models to enhance adaptability or decision efficiency. Other studies integrate MBD-based feature information, case reasoning, semantic knowledge representation, and learning-based prediction within a unified framework (). Such combinations are particularly relevant for complex structural parts, mixed-variety production, and disturbance-prone environments, where purely rule-based or purely data-driven approaches are often insufficient.

The main value of hybrid approaches lies in their balance. Compared with knowledge-driven methods, they are more adaptive; compared with purely optimization-based methods, they can better incorporate process semantics and domain constraints; and compared with purely data-driven methods, they often provide stronger interpretability and more stable decision boundaries. For this reason, hybrid approaches are increasingly viewed as a practical direction for industrially oriented intelligent process planning (), particularly in scenarios requiring both adaptability and constraint-aware decision-making ().

At the same time, hybridization does not automatically guarantee practical success. The main bottlenecks now lie in unified representation, module coordination, and system integration. In particular, knowledge, data, and optimization models are often developed in separate technical pipelines, which makes real-time coordination difficult. Overall, the four methodological paradigms differ in emphasis, but their development trend is increasingly converging toward deeper knowledge–data–algorithm fusion. From a review perspective, this convergence is one of the clearest signs that intelligent process planning is moving from isolated method development toward system-level intelligence.

A key observation across these paradigms is that their differences are not limited to implementation techniques, but also reflect distinct planning logics. Knowledge-driven methods emphasize interpretability and explicit constraint handling; optimization-based methods focus on route quality under multi-objective trade-offs; data-driven methods offer stronger adaptability but depend heavily on data quality and model trustworthiness; and hybrid approaches attempt to balance these strengths at the cost of greater system complexity. This also explains why recent studies increasingly move from standalone methods toward integrated planning architectures.

3 Enabling technologies for intelligent process planning

These technologies provide the data foundation, semantic structure, and feedback capability required for practical deployment. In this sense, they form the infrastructure that enables intelligent process planning to move from conceptual models to engineering implementation.

3.1 Foundational technologies for data and knowledge support

3.1.1 Model-based definition

Model-based definition plays a foundational role in intelligent process planning because it provides a unified carrier for design and manufacturing information. In traditional workflows, geometric data, tolerances, surface requirements, and material information are often distributed across separate drawings and models, which makes downstream planning dependent on incomplete interpretation and manual conversion (). This fragmentation limits the accuracy and consistency of process planning.

By contrast, MBD integrates product geometry and manufacturing-related attributes into a single digital model. This improves information completeness and reduces the loss of critical process data during transmission from design to manufacturing (). For intelligent planning, the importance of MBD lies less in visualization itself and more in the fact that it provides structured, machine-readable information for feature recognition, process reasoning, and route generation. Existing studies have shown that MBD can support automatic extraction of geometric and process information, facilitate machining feature recognition, and improve the consistency of route planning across different part types (; ). In this respect, MBD should be understood as an information infrastructure for process intelligence. Its main contribution is to make planning inputs more complete, standardized, and computable, thereby reducing one of the most persistent bottlenecks in intelligent process planning: unreliable upstream information.

3.1.2 Knowledge graphs and semantic technologies

If MBD addresses the problem of structured product information, knowledge graphs and semantic technologies address the problem of fragmented process knowledge. In machining process planning, knowledge is often scattered across expert experience, handbooks, enterprise rules, historical cases, and production records. Without semantic organization, such knowledge remains difficult to retrieve, reuse, and reason over.

Knowledge graphs provide a structured way to represent process knowledge through explicit entities, relationships, and constraints. This improves semantic association among process resources, part features, operations, machines, and planning rules (). As a result, knowledge graphs can support process recommendation, resource matching, and reasoning-based decision-making more effectively than conventional isolated rule sets or document-based knowledge storage. Their significance in intelligent process planning lies in two aspects. First, they improve knowledge accessibility by transforming dispersed process experience into a machine-readable structure. Second, they create the conditions for more advanced reasoning and interaction, especially when combined with learning models or large language models (). However, their limitations are also clear: knowledge acquisition remains costly, semantic consistency is difficult to maintain, and dynamic updating is still challenging in real industrial environments (). Therefore, knowledge graphs are highly promising, but their practical value depends on whether they can be maintained as living knowledge systems rather than static repositories.

3.2 Technologies for dynamic optimization and closed-loop adaptation

Digital twin technology has become increasingly important because traditional process planning is still dominated by offline decision-making. Once a route is generated, it is often difficult to adjust it efficiently in response to machine failure, material delay, urgent order insertion, or process deviation (). This limitation is particularly serious in multi-variety and small-batch production, where planning conditions may change frequently.

Digital twins provide a pathway toward dynamic and closed-loop process planning by linking physical manufacturing systems with virtual models through continuous data exchange. In this framework, workshops, equipment, operations, and process states can be represented, monitored, and evaluated in a synchronized digital environment (). This enables process routes to be updated not only on the basis of prior knowledge or static optimization, but also on the basis of real-time operating conditions.

The role of digital twins in intelligent process planning should therefore be viewed as more than a visualization tool. Their real value lies in supporting closed-loop optimization, real-time disturbance response, and adaptive reconfiguration. When integrated with MBD and knowledge graph technologies, digital twins can connect upstream information models with downstream execution feedback, thereby forming a more complete chain of perception, reasoning, simulation, and adjustment (). This makes them a key enabling technology for moving intelligent process planning from offline assistance to online decision support.

3.3 Integrative perspective on enabling technologies

Taken together, the enabling technologies discussed above support intelligent process planning at different but complementary levels. MBD improves the completeness and computability of product and process information; knowledge graphs strengthen the organization, sharing, and reasoning of process knowledge; and digital twins enhance real-time feedback and dynamic adaptation. Their functions are therefore not redundant. Instead, they correspond to three essential requirements of intelligent process planning: reliable information input, structured knowledge support, and closed-loop operational feedback.

From this perspective, future progress in intelligent process planning will depend not only on better algorithms, but also on how effectively these enabling technologies are integrated into a unified planning architecture. This is precisely why current research is increasingly shifting from isolated technical improvements toward more coherent, system-level solutions.

The scope of this mini-review should also be interpreted accordingly. Because the purpose is to compare representative methodological paradigms rather than to conduct a full systematic review or meta-analysis, the paper emphasizes reproducible search terms, explicit screening criteria, and transparent selection logic, but it does not claim exhaustive coverage of all CAPP-related publications.

4 Discussion

4.1 Existing problems

Despite considerable progress, intelligent machining process planning still faces several obstacles that limit its broader industrial application. The main difficulty is that current studies have not yet fully bridged the gap between methodological development and practical deployment.

On the one hand, process knowledge remains heterogeneous, experience-dependent, and difficult to formalize, which constrains knowledge reuse, reasoning efficiency, and interpretability. On the other hand, existing optimization and data-driven methods still show limited robustness under dynamic manufacturing conditions, where multiple objectives, changing constraints, and uncertain disturbances must be handled simultaneously (). In addition, the integration of knowledge, data, and optimization is still insufficient, and many studies remain confined to isolated models or laboratory-scale validation rather than real production environments. As a result, challenges in system interoperability, scalability, and evaluation standards continue to hinder the transition of intelligent process planning from theoretical research to industrial practice. Benchmark-level evidence comparing reinforcement learning with classical metaheuristics in manufacturing optimization remains limited, which also weakens the generalizability of performance claims (). Because direct numerical comparison across studies remains difficult due to differences in planning tasks, benchmark settings, and objective formulations, Table 2 summarizes representative studies together with the evaluation metrics explicitly reported or emphasized in the literature.

TABLE 2

StudyMethod categoryPlanning taskReported or emphasized metric(s)Main reported resultComparability note
Knowledge graphAutomatic machining process decision-makingDecision time; reasoning order; comparison with ontology/database methodsKG-based decision-making was reported to improve reasoning efficiency for complex part-level decisionsPrototype and case-specific evaluation; not a unified benchmark
Multi-objective optimizationSmart crankshaft process planningMachining and non-machining time; tool-life-related cost; surface roughness/product qualityMulti-objective planning was used to automate process parameters, operation sequencing, and turret tool positioningFocused on crankshaft machining; objective weights and constraints are application-specific
Optimization-basedProcess route optimizationConvergence behavior; solution quality; route optimization performanceAdaptive improved ant colony optimization was reported to improve route optimization under the tested settingBenchmark setting and objective formulation differ from other studies
Deep reinforcement learningDynamic process planning with changing resourcesDecision/response speed; operation sequencing; resource selection feasibilityDRL was used to support fast decision-making under dynamic machining-resource conditionsDifficult to compare directly with static metaheuristics because resource states are dynamic
Data-driven + knowledge-guidedNC machining process planningPlanning effectiveness; knowledge-guided recommendation; model performanceData-driven learning was combined with process knowledge to support NC process planningReported metrics are tied to the authors task definition; public benchmark comparability is limited
Knowledge graph + DRLIntelligent generation of machining process designCase-study validation; process-generation effectiveness; reasoning reuseKG and DRL were combined to support unified representation and intelligent generation of process designValidated through a specific component case; cross-part generalization needs further testing
Hybrid reasoningGrinding process intelligent decisionDecision effectiveness; case retrieval/adaptation performanceCase-based reasoning and process reasoning improved intelligent decision support for typical grinding partsCase-specific domain; limited evidence for broader machining-route transfer
Digital-twin-enabledMachine-tool intelligence and adaptive supportSystem responsiveness; adaptive capability; feedback integrationDigital-twin modeling supports real-time feedback and adaptive decision supportNot a direct route-planning algorithm benchmark
RL vs. metaheuristicsManufacturing optimization comparisonComparative performance metrics under benchmark settingsComparison outcomes depend strongly on benchmark design and problem settingDifferent manufacturing-optimization domain; useful mainly as cautionary benchmark evidence

Representative studies and reported evaluation metrics in intelligent machining process planning.

4.2 Current controversies

Beyond these systemic challenges, several key technologies still involve controversies that require in-depth resolution:

  • Application risks of large language models (LLMs): Although LLMs show promise in semantic interaction and process recommendation in manufacturing and process planning, their practical application in computer-aided process planning (CAPP) remains uncertain (; ). Outputs that suffer from hallucination or violate process constraints may introduce unacceptable risks into machining plans. This concern has motivated recent attempts to combine LLMs with knowledge-graph-based grounding mechanisms ().

  • Long-term maintainability of knowledge graphs: While knowledge graphs provide a powerful framework for structured reasoning, their long-term scalability in real industrial environments depends heavily on the cost of knowledge acquisition, updating, sharing, and maintenance (). Recent manufacturing knowledge systems further suggest that sustained value creation requires continuous capture of both explicit and implicit knowledge, which increases implementation complexity in practice ().

  • Generalizability of reinforcement learning (RL)-based planning: RL-based planning is attractive for dynamic decision-making, but its reported advantages over classical metaheuristics remain difficult to generalize across studies and application settings. Across manufacturing optimization studies, benchmark settings, objective functions, and evaluation metrics vary substantially, which undermines the credibility of direct performance comparisons (; ). Recent reviews of RL in production scheduling also indicate that cross-study inferential consistency remains limited ().

4.3 Future development trends

Future research on intelligent machining process planning is expected to evolve toward a more integrated, adaptive, and application-oriented paradigm. Rather than relying on isolated improvements in individual methods, the next stage of development will likely emphasize the coordinated advancement of knowledge-enhanced intelligence, closed-loop adaptive optimization, and industrially deployable sustainable systems.

  • Knowledge-enhanced and interpretable intelligent planning will become a major direction.

Future process planning systems are expected to move beyond conventional rule-based or purely data-driven approaches toward deeper integration of knowledge graphs, large language models, and data-driven learning. Such a framework can improve process knowledge extraction, semantic reasoning, dynamic knowledge updating, and human-understandable decision support, thereby enhancing both intelligence and interpretability. However, scalable maintenance of process knowledge graphs and hallucination control in LLM-assisted planning remain open challenges (

).

  • 2. Adaptive closed-loop optimization will be increasingly important in dynamic manufacturing environments.

With the growing complexity of multi-variety and small-batch production, process planning must gradually shift from static offline optimization to real-time, multi-objective, and context-aware decision-making. In this context, the integration of lightweight optimization algorithms with digital twin technology will provide a promising pathway for achieving virtual–real interaction, dynamic response, and closed-loop optimization under changing production conditions.

  • 3. System-level integration and sustainable deployment will be critical for industrial application.

Future development should place greater emphasis on interoperability among CAD, CAM, MES, ERP, and shop-floor systems, so that process planning can be embedded into the broader manufacturing workflow. At the same time, green indicators such as energy consumption, carbon emissions, and resource utilization should be incorporated into planning objectives, enabling intelligent process planning to progress toward scalable, sustainable, and industrially applicable solutions.

5 Conclusion

Intelligent machining process planning has become an essential component of intelligent manufacturing because it plays a decisive role in production efficiency, cost control, and product quality. This review shows that the field has evolved from conventional experience-based planning toward knowledge-driven, optimization-based, data-driven, and hybrid intelligent paradigms. Although substantial progress has been achieved in multi-objective optimization, knowledge representation, and the integration of enabling technologies such as model-based definition, knowledge graphs, and digital twins, important challenges remain in interpretability, robustness, cross-system integration, and industrial deployment. Overall, the future of intelligent machining process planning will depend less on isolated improvements in individual methods and more on the development of integrated, adaptive, and practically deployable frameworks. Advancing along this direction will be critical for translating intelligent process planning from academic research into scalable manufacturing practice.

Statements

Author contributions

LC: Data curation, Writing – original draft. LZ: Funding acquisition, Supervision, Writing – review and editing. MZ: Resources, Writing – review and editing.

Funding

The author(s) declared that financial support was received for this work and/or its publication. The authors declare that the research is supported by Specialized Intelligent Manufacturing Engineering Laboratory, Platform ID: PT2025KJS002.

Conflict of interest

The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

References

  • 1

    AfifM. F.SarhanA. A. D. (2025). Computer-aided process planning, digital twin, and smart manufacturing: interconnections and integration in CNC machining processes. Int. J. Prod. Res.63, 95939632. 10.1080/00207543.2025.2521717

  • 2

    AhmedM. J.MozoA.KaramchandaniA. (2025). A survey on graph neural networks, machine learning and deep learning techniques for time series applications in industry. PeerJ Comput. Sci.11, e3097. 10.7717/peerj-cs.3097

  • 3

    AmouzgarK.WangW.EynianM.NgA. H. (2025). Smart process planning of crankshaft machining through multiple objectives optimization. Procedia CIRP134, 241246. 10.1016/j.procir.2025.03.018

  • 4

    AsanoT.WatanabeY.NakamotoK. (2022). Proposal of a machining feature recognition method to reflect product and manufacturing information. J. Adv. Mech. Des. Syst. Manuf.16, JAMDSM0015. 10.1299/jamdsm.2022jamdsm0015

  • 5

    Besharati-FoumaniH.LohtanderM.VarisJ. (2019). Intelligent process planning for smart manufacturing systems: a state-of-the-art review. Procedia Manuf.38, 156162. 10.1016/j.promfg.2020.01.021

  • 6

    ElbasheerM.LailiY.LongoF.SolinaV.TaoY.VeltriP.et al (2025). Natural language-driven production planning: integrating large language models with automatic simulation model generation in manufacturing systems. J. Intell. Manuf.10.1007/s10845-025-02732-z

  • 7

    FanH.FanZ.LiuC.ZhuJ.GibbsT.FuhJ. Y. H.et al (2025). MetalMind: a knowledge graph-driven human-centric knowledge system for metal additive manufacturing. Npj Adv. Manuf.2, 25. 10.1038/s44334-025-00038-9

  • 8

    FuY.WangY.GaoK.HuangM. (2024). Review on ensemble meta-heuristics and reinforcement learning for manufacturing scheduling problems. Comput. Electr. Eng.120, 109780. 10.1016/j.compeleceng.2024.109780

  • 9

    GuoL.YanF.LuY.ZhouM.YangT. (2021). An automatic machining process decision-making system based on knowledge graph. Int. J. Comput. Integr. Manuf.34, 13481369. 10.1080/0951192X.2021.1972461

  • 10

    GuoL.HuT.DongL.MaS. (2024a). Ontology and production rules-based dynamic knowledge base construction methodology for machining process. J. Manuf. Syst.77, 10271044. 10.1016/j.jmsy.2024.11.006

  • 11

    GuoL.LiX.YanF.LuY.ShenW. (2024b). A method for constructing a machining knowledge graph using an improved transformer. Expert Syst. Appl.237, 121448. 10.1016/j.eswa.2023.121448

  • 12

    HanZ.HuangR.HuangB.JiangJ.LiX. (2023). Data-driven and knowledge-guided approach for NC machining process planning. Computer-Aided Des.162, 103562. 10.1016/j.cad.2023.103562

  • 13

    HollandM.ChaudhariK. (2024). Large language model based agent for process planning of fiber composite structures. Manuf. Lett.40, 100103. 10.1016/j.mfglet.2024.03.010

  • 14

    HuaY.WangR.WangZ.WangG.YanY. (2025). Knowledge graph with deep reinforcement learning for intelligent generation of machining process design. J. Eng. Des.36, 20722106. 10.1080/09544828.2024.2338342

  • 15

    HuangL.YuW.MaW.ZhongW.FengZ.WangH.et al (2025). A survey on hallucination in large language models: principles, taxonomy, challenges, and open questions. ACM Trans. Inf. Syst.43 (42), 155. 10.1145/3703155

  • 16

    ImadM.HopkinsC.HosseiniA.YussefianN. Z.KishawyH. A. (2022). Intelligent machining: a review of trends, achievements and current progress. Int. J. Comput. Integr. Manuf.35, 359387. 10.1080/0951192X.2021.1891573

  • 17

    ItuA. (2025). Industrial scheduling in the digital era: challenges, state-of-the-art methods, and deep learning perspectives. Appl. Sci.15, 10823. 10.3390/app151910823

  • 18

    KesslerM.ArlinghausJ. C. (2022). A framework for human-centered production planning and control in smart manufacturing. J. Manuf. Syst.65, 220232. 10.1016/j.jmsy.2022.09.013

  • 19

    KlarM.GlattM.AurichJ. C. (2023). Performance comparison of reinforcement learning and metaheuristics for factory layout planning. CIRP J. Manuf. Sci. Technol.45, 1025. 10.1016/j.cirpj.2023.05.008

  • 20

    KusyiY.OnyskoO.KukA.KostiukO.SolohubB. (2023). “Improvement of the technique of optimal technological routes planning for machining of the machine parts surfaces,” in New Technologies, Development and Application VI. Editors KarabegovicI.KovačevićA.MandzukaS. (Cham: Springer Nature Switzerland), 221232. 10.1007/978-3-031-31066-9_23

  • 21

    LazaroiuG.AndroniceanuA.GrecuI.GrecuG.NegurițăO. (2022). Artificial intelligence-based decision-making algorithms, internet of things sensing networks, and sustainable cyber-physical management systems in big data-driven cognitive manufacturing. Oeconomia Copernic.13, 10471080. 10.24136/oc.2022.030

  • 22

    LiY.ZhouT. (2025). Research on intelligent planning method for turning machining process based on knowledge base. Machines13, 417. 10.3390/machines13050417

  • 23

    LiZ.DengZ.GeZ.LvL.GeJ. (2023). A hybrid approach of case-based reasoning and process reasoning to typical parts grinding process intelligent decision. Int. J. Prod. Res.61, 503519. 10.1080/00207543.2021.2010144

  • 24

    LiJ.ZhouG.ZhangC.HuJ.ChangF.MattaA. (2025). Defining a feature-level digital twin process model by extracting machining features from MBD models for intelligent process planning. J. Intelligent Manuf.36, 32273248. 10.1007/s10845-024-02406-2

  • 25

    MarziaS.AzabA. (2023). Automated process planning and dynamic scheduling for smart manufacturing: a systematic literature review. Manuf. Lett.35, 861872. 10.1016/j.mfglet.2023.07.013

  • 26

    ModrakV.SudhakarapandianR.BalamuruganA.SoltysovaZ. (2024). A review on reinforcement learning in production scheduling: an inferential perspective. Algorithms17, 343. 10.3390/a17080343

  • 27

    RahmaniS.AghalarH.JebreiliS.GoliA. (2024). “Optimization and computing using intelligent data-driven approaches for decision-making,” in Optimization and Computing Using Intelligent data-driven Approaches for decision-making (Boca Raton, FL: CRC Press), 90176. Available online at: https://api.taylorfrancis.com/content/chapters/edit/download?identifierName=doi&identifierValue=10.1201/9781003536796-6&type=chapterpdf (Accessed March 30, 2026).

  • 28

    SherifZ.SalonitisK. (2025). A systematic review of decision tools for process selection and performance improvement in manufacturing. Int. J. Adv. Manuf. Technol.141, 11131141. 10.1007/s00170-025-16806-y

  • 29

    SunP. Z.BaoY.MingX.ZhouT. (2022). Knowledge-driven industrial intelligent system: concept, reference model, and application direction. IEEE Trans. Comput. Soc. Syst.10, 14651478. 10.1109/tcss.2022.3188295

  • 30

    WangB.WuY.-B.WangM.-J. (2025). Process route optimization based on FAHP and adaptive improved ant colony algorithm. J. Adv. Manuf. Syst.24, 735767. 10.1142/S0219686725500325

  • 31

    Waubert De PuiseauC.MeyesR.MeisenT. (2022). On reliability of reinforcement learning based production scheduling systems: a comparative survey. J. Intell. Manuf.33, 911927. 10.1007/s10845-022-01915-2

  • 32

    WenP.MaY.WangR. (2023). Systematic knowledge modeling and extraction methods for manufacturing process planning based on knowledge graph. Adv. Eng. Inf.58, 102172. 10.1016/j.aei.2023.102172

  • 33

    WuZ.LiangC. (2024). A review and prospects of manufacturing process knowledge acquisition, representation, and application. Machines12, 416. 10.3390/machines12060416

  • 34

    WuW.HuangZ.ZengJ.FanK. (2021). A fast decision-making method for process planning with dynamic machining resources via deep reinforcement learning. J. Manufacturing Systems58, 392411. 10.1016/j.jmsy.2020.12.015

  • 35

    XiaoY.ZhengS.ShiJ.DuX.HongJ. (2023). Knowledge graph-based manufacturing process planning: a state-of-the-art review. J. Manuf. Syst.70, 417435. 10.1016/j.jmsy.2023.08.006

  • 36

    XieY.LiuJ.WangR.WangZ.YuK.SongZ. (2025). Rapid generation method of process routes based on multi-agent collaboration with LLMs. Adv. Eng. Inf.68, 103733. 10.1016/j.aei.2025.103733

  • 37

    XuX.WangL.NewmanS. T. (2011). Computer-aided process planning – a critical review of recent developments and future trends. Int. J. Comput. Integr. Manuf.24, 131. 10.1080/0951192X.2010.518632

  • 38

    XuT.LiJ.ChenZ. (2022). Automatic machining feature recognition based on MBD and process semantics. Comput. Industry142, 103736. 10.1016/j.compind.2022.103736

  • 39

    XuQ.QiuF.ZhouG.ZhangC.DingK.ChangF.et al (2025). A large language model-enabled machining process knowledge graph construction method for intelligent process planning. Adv. Eng. Inf.65, 103244. 10.1016/j.aei.2025.103244

  • 40

    YuJ.ZhangY. (2023). Challenges and opportunities of deep learning-based process fault detection and diagnosis: a review. Neural Comput. Appl.35, 211252. 10.1007/s00521-022-08017-3

  • 41

    YunH.KimE.KimD. M.ParkH. W.JunM. B.-G. (2023). Machine learning for object recognition in manufacturing applications. Int. J. Precis. Eng. Manuf.24, 683712. 10.1007/s12541-022-00764-6

  • 42

    YusofY.LatifK. (2014). Survey on computer-aided process planning. Int. J. Adv. Manuf. Technol.75, 7789. 10.1007/s00170-014-6073-3

  • 43

    ZhangL.LiuJ.ZhuangC. (2024a). Digital twin modeling enabled machine tool intelligence: a review. Chin. J. Mech. Eng.37, 47. 10.1186/s10033-024-01036-2

  • 44

    ZhangX.ZhengL.FanW.JiW.MaoL.WangL. (2024b). Knowledge graph and function block based digital twin modeling for robotic machining of large-scale components. Robotics Computer-Integrated Manuf.85, 102609. 10.1016/j.rcim.2023.102609

  • 45

    ZhangG.ZhaoY.WangG.ZhangY. (2026). AI-driven design and optimisation control of service-oriented smart manufacturing systems: applications, challenges, and future directions. J. Eng. Des.37, 156. 10.1080/09544828.2026.2640627

  • 46

    ZhaoX.WeiS.RenS.CaiW.ZhangY. (2024). Integrating MBD with BOM for consistent data transformation during lifecycle synergetic decision-making of complex products. Adv. Eng. Inf.61, 102491. 10.1016/j.aei.2024.102491

  • 47

    ZhouG.HanC.ZhangC.ZhouY.ZengK.LiuJ.et al (2025). Interpretable knowledge recommendation for intelligent process planning with graph embedded deep reinforcement learning. Adv. Eng. Inf.65, 103321. 10.1016/j.aei.2025.103321

Summary

Keywords

hybrid intelligence, intelligent decision-making, knowledge graph, machining process planning, multi-objective optimization

Citation

Chen L, Zheng L and Zhu M (2026) A review of intelligent decision-making and planning methods for machining process routes. Front. Mech. Eng. 12:1842661. doi: 10.3389/fmech.2026.1842661

Received

30 March 2026

Revised

29 June 2026

Accepted

06 July 2026

Published

23 July 2026

Volume

12 - 2026

Edited by

Viet Q. VU, Thai Nguyen University of Technology, Vietnam

Reviewed by

Minh-Quang Tran, TUETECH University, Vietnam

David Alfaro, University of Costa Rica, Costa Rica

Updates

Copyright

*Correspondence: Liancheng Zheng,

Disclaimer

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.

Outline

Figures

Cite article

Copy to clipboard


Export citation file


Share article

Article metrics