Artificial Intelligence in Production and Optimization

Abstract

This review is a systematic literature review-cum-bibliometric analysis of the field of Artificial Intelligence in Production and Optimization with a view to mapping the field in holistic manner. In this review the authors have attempted to identify the predominant strands in the field and identify where learning approaches have been linked to decision making frameworks and where learning approaches end at prediction. Data for the review are sourced from Elsevier\'s Scopus database, and are limited to English language journal articles and reviews published between 2014 and 2027 using key words. This review is based on 3,727 records in the final database. The number of documents published has increased from 38 in 2014 to 1,161 in 2026 and the top producers are China, India, and the United States. The most prevalent themes were identified as Artificial Intelligence, Machine Learning, and Optimization. The formal decision model seems to be mainly in the field of scheduling, while in bioprocess and additive manufacturing, and food systems, it is not often that the decision is made to end at prediction.

Introduction

This study presents a bibliometric and systematic literature review of artificial intelligence (AI) applications in production and optimisation, focusing on whether machine-learning predictions are connected to formal decision-making models. Although AI is increasingly used to improve productivity, quality, energy efficiency and sustainability, accurate predictions do not necessarily lead to optimal operational decisions. The review therefore investigates how AI can be integrated with operations research (OR) techniques, such as linear programming, mixed-integer programming, stochastic optimisation and robust optimisation.

The study analyses Scopus publications from 2014 to 2027 using Bibliometrix/Biblioshiny, VOSviewer and Excel. The search initially identified 8,311 records, which were progressively filtered by publication year, document type, language and source type to obtain 3,727 journal articles and reviews. The performance dataset contains 3,725 documents from 1,484 sources, while the review corpus contains 3,719 unique records. Publication output increased sharply in recent years, with more than half of the documents dated 2025 or 2026. However, the 2026 figures are incomplete because the data were exported on 11 September 2026. The analysis examines publication trends, citations, authorship, countries, journals, keywords and research collaboration. Because cited references were unavailable in the export, co-citation and bibliographic coupling analyses could not be performed.

The findings indicate that AI research in production and optimisation is expanding rapidly across manufacturing, renewable energy, bioprocessing, additive manufacturing, agriculture and food supply chains. Nevertheless, the integration of machine learning with formal optimisation remains limited. Scheduling is the clearest area of integration, where reinforcement learning and other learning-based approaches support production planning, job-shop scheduling and rescheduling. Some studies combine machine-learning forecasts with linear programming models, allowing predictions to inform decisions while mathematical optimisation ensures feasibility.

In contrast, predictive maintenance, quality control and digital twins often focus on predicting failures, defects or process outcomes without translating these predictions into maintenance schedules, resource allocations or production plans. Similar limitations appear in fermentation, biofuel production, additive manufacturing and food systems. For example, machine learning may predict biomass yield or product printability, but the subsequent decisions about production capacity, scheduling and resource use are often not formally optimised.

The review identifies four major research gaps:

1. Limited integration of AI and formal optimisation: Learning-based decision models are concentrated in scheduling, with relatively little application to bioprocesses, additive manufacturing and food systems.

2. Inadequate treatment of uncertainty: Robust and stochastic optimisation methods remain uncommon, despite the uncertainty associated with demand, resources, equipment failures and environmental conditions.

3. Insufficient integration of explainability and human oversight: Interpretability, human–AI collaboration and privacy-preserving methods are often studied separately rather than incorporated into operational decision models.

4. Neglect of AI's environmental costs: Energy consumption and the environmental footprint of AI systems are rarely included directly in optimisation objectives, even though their impacts can vary substantially across regions.

The review also highlights emerging research areas, including digital twins, Industry 5.0, federated learning, large language models and quantum-assisted optimisation. These developments offer opportunities to build more adaptive and intelligent production systems, but require stronger connections between predictive capabilities and practical operational decisions.

Conclusion

The four questions have a single answer. Output was small from 2014 to 2018 and then rose sharply, with more than half of the documents dated 2025 or 2026. China accounts for about 30 per cent of the file, and authorship is concentrated in a few surname-and-initial groups. The keyword maps place artificial intelligence in the centre, with industry and decisions, energy and environment, life sciences, and optimisation methods around it. Formal decision models appear mainly in scheduling (Esteso et al., 2023; Zhang et al., 2023). In bioprocess, additive manufacturing, and food systems, the published work more often stops at a prediction (Elbadawi et al., 2020; Kollia et al., 2021; Murali et al., 2026). The performance file makes the growth concrete without supporting a forecast. The 3,725 documents come from 1,484 sources and 12,416 authors. Only 222 authors wrote alone. The mean is 4.74 co-authors per document, and international co-authorship is 31.17 per cent. The file holds 11,142 author keywords. Of the documents, 2,709 are articles and 1,016 are reviews. The mean citation count is 21.43, and the total is 79,822. The reported growth rate of ?6.44 per cent is an artefact of the window. It compares the first year with the last, and 2027 has 16 documents. Set 2014 against 2026 and the rise is about 33 per cent a year. The 2026 count will finish higher than 1,161, because the export is dated 11 September 2026. A linear trend line should not be used on this series. The practical reading is that the field is young, written in teams, and still filling in. The average document age is 2.23 years. A bibliometric reading is suited to a set of this size precisely because a narrative review cannot cover it (Donthu et al., 2021). Citations are headed by one paper and otherwise spread. (Tao & Zhang, 2017) has 1,385 citations. Applied Sciences leads the sources with 75 documents, then Sustainability, IEEE Access, the International Journal of Advanced Manufacturing Technology, and Energies, with 62, 61, 52, and 50. Even the top journal holds about 2 per cent. A reader who follows one outlet, or the single most cited paper, will miss most of the set. That scatter is why a systematic selection was used (Linnenluecke et al., 2020; Paul et al., 2021). Country counts are documents with at least one author from that country, so a two-country paper counts twice. China has 1,099 documents, India 532, and the United States 359. Saudi Arabia, Iran, the United Kingdom, South Korea, Spain, Italy, and Malaysia follow, with 213, 189, 174, 156, 134, 133, and 130. Six of the top ten are in Asia. Grey areas on the map, including Russia, Greenland, and parts of central Africa, mean little or no data in this file. The work is international, and a few large countries still produce most of it. The three-field plot joins the top 15 countries, the top 20 authors, and the top 15 keywords. China is the largest country box. Almost all of the top author labels are Chinese surnames with an initial: Wang Y (80 documents), Zhang Y (75), Wang X (57), Li Y (55), and Liu Y (54). Scopus stores authors as surname and initial, so these are name groups, not single researchers. Artificial intelligence is the largest keyword box, with 1,206 documents, then machine learning with 566 and optimization with 236. The file counts “artificial intelligence” and “artificial intelligence (ai)” apart, which splits one concept across two nodes. The VOSviewer maps agree on the structure and disagree only on how finely it is cut (van Eck and Waltman, 2010). At 30 occurrences there are four large clusters and one small one. Artificial intelligence links to every cluster. One cluster covers factory and decision topics, including digital twin, industry 4.0, smart manufacturing, and scheduling. One covers energy and environment, including hydrogen, renewable energy, biomass, and sustainability. One holds optimisation methods, including genetic algorithms, particle swarm optimization, and neural networks. One holds life-science indexing terms, including metabolism, biosynthesis, and drug development. A small group covers crops. At 15 occurrences the same core splits, and small niches appear, including oil fields, wind turbines, and data privacy. A simpler network of about 45 keywords falls into three clusters: optimisation with energy, learning with smart factories, and study terms with life-science words. The maps are a picture of vocabulary, not of decision quality (Zupic and ?ater, 2015). A large node means a word is used often. It does not mean the papers under that word solve an allocation, a schedule, or a capacity problem. The review looks past those large nodes. In the 3,719-record corpus, additive manufacturing has 58 records, food safety 24, microalgae 41, and anaerobic digestion 63. Digital twin has 135, predictive maintenance 89, and reinforcement learning 107. Robust optimization has 9 and vehicle routing has 4. (Esteso et al., 2023) places that learning mainly in simplified scheduling environments, so the advantage over a tuned exact solver is not established. (Y. Zhang et al., 2022) and (Li et al., 2020) sit on the same scheduling line, as do (Usman et al., 2026) and (Prata et al., 2024). Cases that keep an optimisation model and use learning only to forecast cost or demand remain the exception (Boresta et al., 2024); (Vaz et al., 2024). Robust treatments are real but rare (Ajagekar et al., 2023; R. Zhang et al., 2020). (Danach et al., 2026) is one of the few records that carries a forecast into a schedule. (Wahid et al., 2022) and (Christou et al., 2022) stop earlier, at the failure forecast and the quality platform. The bioprocess and small-circle records confirm the split. Two records out of 418 on fermentation, metabolism, microalgae, or bioprocesses mention a formal programming method or model predictive control. (Imamoglu, 2024) and (Hossain et al., 2022) model productivity. (Kabir et al., 2023) models hydrogen yield. (Murali et al., 2026) makes the plant-level gap explicit. (Soori et al., 2025) and (Ciccone et al., 2023) keep additive manufacturing on quality, design, and materials. (Elbadawi et al., 2020) predicts printability and does not schedule the queue. (Serov & Vinogradov, 2022) still treats nanomedicine design as early. Food-chain studies detect contamination or forecast yield (Dakhia et al., 2025; Kollia et al., 2021; Pandey et al., 2023) and do not route the fleet. (Tsolakis et al., 2022) is the logistics exception, and it notes the missing environmental term in the decision model. Four gaps follow, and they are the agenda. First, learning should be tested against an exact or heuristic benchmark, not only against another learning policy, and the test should use a realistic instance (Esteso et al., 2023). Scheduling is where that comparison is easiest to run. Bioprocess, additive manufacturing, and food systems are where it is still rare (Murali et al., 2026; Ciccone et al., 2023; Kollia et al., 2021), which is why a single well-specified study in one of those domains would move the field more than another scheduling variant on a toy shop. Second, uncertainty needs a formulation. Nine records on robust optimization, against 107 on reinforcement learning, is not a balance that matches plant data (Ajagekar et al., 2023). Third, explainability, human oversight, and privacy-preserving training should enter the decision model as constraints, not as a separate keyword (Danach et al., 2026; Nguyen et al., 2025; Tóth et al., 2023; Yang et al., 2022). Fourth, the objective should include the resource cost of the model. (Wheeler, 2026), drawing on 73 empirical studies, reports that environmental outcomes vary tenfold to sixtyfold across regions. A method that ignores that cost will rank actions differently in different places. Newer tools do not close these gaps by themselves. (Chen & Huang, 2025), (Bo?ejko et al., 2026), and (Ajagekar& You, 2022) extend the method set. They do not yet supply the missing plant-level decision models. Several limits follow from the design, and they bound what the agenda can claim. The search used one database and one Boolean string (Donthu et al., 2021). The filters removed 879 documents on time, 3,298 on document type, 379 on language, and 28 on journal. Conference papers are outside the set. The export has no cited references, so co-citation could not be run, and the intellectual base of the field is not identified here (Zupic and ?ater, 2015). Author counts for Wang Y and similar labels are name groups. Keyword variants were left split. The 2026 total is incomplete, and the 2027 count should not be read as decline. Country totals double-count internationally co-authored papers. The review corpus, the performance file, and the journal-filter total differ by a small margin: 3,719, 3,725, and 3,727. PRISMA records the selection path. It does not appraise the quality of each paper (Moher et al., 2010). What the file supports is a description, not a ranking of methods. The literature is young, team-based, and Asia-heavy, spread across 1,484 sources, with more than half of its documents dated 2025 or 2026. The maps put artificial intelligence, machine learning, and optimization at the centre, and energy, smart manufacturing, and life sciences around them. The review shows that the small circles are where a formal decision model is still rare (Murali et al., 2026; Soori et al., 2025). The next useful paper is not another map of the same three words. It is a study that recommends an action in one of those circles, states the uncertainty, and tests the recommendation against a decision model.

References

[1] Ajagekar, A. and You, F. (2022) Quantum computing and quantum artificial intelligence for renewable and sustainable energy: a emerging prospect towards climate neutrality. Renewable and Sustainable Energy Reviews, 165. https://doi.org/10.1016/j.rser.2022.112493 [2] Ajagekar, A., Mattson, N.S. and You, F. (2023) Energy-efficient AI-based control of semi-closed greenhouses leveraging robust optimization in deep reinforcement learning. Advances in Applied Energy, 9. https://doi.org/10.1016/j.adapen.2022.100119 [3] Boresta, M., Pinto, D.M. and Stecca, G. (2024) Bridging operations research and machine learning for service cost prediction in logistics and service industries. Annals of Operations Research, 342(1). https://doi.org/10.1007/s10479-024-05962-1 [4] Bo?ejko, W., Trotskyi, S., Uchro?ski, M. and Wodecki, M. (2026) Optimizing two-machine scheduling in flexible manufacturing systems using autonomous AI and quantum computing. Neurocomputing, 664. https://doi.org/10.1016/j.neucom.2025.132067 [5] Chen, D. and Huang, Y. (2025) Integrating reinforcement learning and large language models for crop production process management optimization and control through a new knowledge-based deep learning paradigm. Computers and Electronics in Agriculture, 232. https://doi.org/10.1016/j.compag.2025.110028 [6] Chen, S.C., Chen, H.M., Chen, H.K. and Li, C.L. (2024) Multi-objective optimization in Industry 5.0: human-centric AI integration for sustainable and intelligent manufacturing. Processes, 12(12). https://doi.org/10.3390/pr12122723 [7] Christou, I.T., Kefalakis, N., Soldatos, J.K. and Despotopoulou, A.M. (2022) End-to-end industrial IoT platform for Quality 4.0 applications. Computers in Industry, 137. https://doi.org/10.1016/j.compind.2021.103591 [8] Ciccone, F., Bacciaglia, A. and Ceruti, A. (2023) Optimization with artificial intelligence in additive manufacturing: a systematic review. Journal of the Brazilian Society of Mechanical Sciences and Engineering, 45(6). https://doi.org/10.1007/s40430-023-04200-2 [9] Dakhia, Z., Russo, M. and Merenda, M. (2025) AI-enabled IoT for food computing: challenges, opportunities, and future directions. Sensors, 25(7). https://doi.org/10.3390/s25072147 [10] Danach, K., Harb, H., Issa, H. and Saker, L. (2026) An explainable data-driven optimization framework for industrial predictive maintenance scheduling. Results in Engineering, 29. https://doi.org/10.1016/j.rineng.2026.109022 [11] Elbadawi, M., Muñiz Castro, B., Gavins, F.K.H., Ong, J.J., Gaisford, S., Pérez, G. et al. (2020) M3DISEEN: a novel machine learning approach for predicting the 3D printability of medicines. International Journal of Pharmaceutics, 590. https://doi.org/10.1016/j.ijpharm.2020.119837 [12] Esteso, A., Peidro, D., Mula, J. and Díaz-Madroñero, M. (2023) Reinforcement learning applied to production planning and control. International Journal of Production Research, 61(16). https://doi.org/10.1080/00207543.2022.2104180 [13] Hossain, S.M.Z., Sultana, N., Razzak, S.A. and Hossain, M.M. (2022) Modeling and multi-objective optimization of microalgae biomass production and CO2 biofixation using hybrid intelligence approaches. Renewable and Sustainable Energy Reviews, 157. https://doi.org/10.1016/j.rser.2021.112016 [14] Imamoglu, E. (2024) Artificial intelligence and/or machine learning algorithms in microalgae bioprocesses. Bioengineering, 11(11). https://doi.org/10.3390/bioengineering11111143 [15] Kabir, M.M., Roy, S.K., Alam, F., Nam, S.Y., Im, K.S., Tijing, L. et al. (2023) Machine learning-based prediction and optimization of green hydrogen production technologies from water industries for a circular economy. Desalination, 567. https://doi.org/10.1016/j.desal.2023.116992 [16] Kamran, M.A., Kia, R., Goodarzian, F. and Ghasemi, P. (2023) A new vaccine supply chain network under COVID-19 conditions considering system dynamic: artificial intelligence algorithms. Socio-Economic Planning Sciences, 85. https://doi.org/10.1016/j.seps.2022.101378 [17] Kollia, I., Stevenson, J. and Kollias, S. (2021) AI-enabled efficient and safe food supply chain. Electronics, 10(11). https://doi.org/10.3390/electronics10111223 [18] Li, Y., Carabelli, S., Fadda, E., Manerba, D., Tadei, R. and Terzo, O. (2020) Machine learning and optimization for production rescheduling in Industry 4.0. International Journal of Advanced Manufacturing Technology, 110(9–10). https://doi.org/10.1007/s00170-020-05850-5 [19] Martini, B., Bellisario, D. and Coletti, P. (2024) Human-centered and sustainable artificial intelligence in Industry 5.0: challenges and perspectives. Sustainability, 16(13). https://doi.org/10.3390/su16135448 [20] Murali, R., Bywater, A., Dolat, M., Dekhici, B., Zarei, M., Hilton, L. et al. (2026) Anaerobic digestion site-wide optimisation and decision-making: an industrial perspective and review. Renewable and Sustainable Energy Reviews, 226. https://doi.org/10.1016/j.rser.2025.116402 [21] Ng, W.L., Goh, G.L., Goh, G.D., Ten, J.S.J. and Yeong, W.Y. (2024) Progress and opportunities for machine learning in materials and processes of additive manufacturing. Advanced Materials, 36(34). https://doi.org/10.1002/adma.202310006 [22] Nguyen, Q.T., Truong, T.C., Nguyen, T.P., Tran, T.N., Mesicek, J. and Pagac, M. (2025) Explainable artificial intelligence in additive manufacturing: a systematic review on methods. [journal, volume and DOI missing] [23] Pandey, V.K., Srivastava, S., Dash, K.K., Singh, R., Mukarram, S.A., Kovács, B. et al. (2023) Machine learning algorithms and fundamentals as emerging safety tools in preservation of fruits and vegetables: a review. Processes, 11(6). https://doi.org/10.3390/pr11061720 [24] Prata, B.A., Abreu, L.R. and Nagano, M.S. (2024) Applications of constraint programming in production scheduling problems: a descriptive bibliometric analysis. Results in Control and Optimization, 14. https://doi.org/10.1016/j.rico.2023.100350 [25] Serov, N. and Vinogradov, V. (2022) Artificial intelligence to bring nanomedicine to life. Advanced Drug Delivery Reviews, 184. https://doi.org/10.1016/j.addr.2022.114194 [26] Soori, M., Jough, F.K.G., Dastres, R. and Arezoo, B. (2025) Additive manufacturing modification by artificial intelligence, machine learning, and deep learning: a review. Additive Manufacturing. [volume and DOI missing] [27] Tao, F. and Zhang, M. (2017) Digital twin shop-floor: a new shop-floor paradigm towards smart manufacturing. IEEE Access, 5. https://doi.org/10.1109/ACCESS.2017.2756061 [28] Tóth, A., Nagy, L., Kennedy, R., Bohuš, B., Abonyi, J. and Ruppert, T. (2023) The human-centric Industry 5.0 collaboration architecture. MethodsX, 11. https://doi.org/10.1016/j.mex.2023.102260 [29] Tsolakis, N., Zissis, D., Papaefthimiou, S. and Korfiatis, N. (2022) Towards AI driven environmental sustainability: an application of automated logistics in container port terminals. International Journal of Production Research, 60(14). https://doi.org/10.1080/00207543.2021.1914355 [30] Ukwaththa, J., Herath, S. and Meddage, D.P.P. (2024) A review of machine learning (ML) and explainable artificial intelligence (XAI) methods in additive manufacturing (3D printing). Materials Today Communications, 41. https://doi.org/10.1016/j.mtcomm.2024.110294 [31] Usman, S., Ye, T., Ahmed, F., Huang, S., Haider, M.H., Qin, W. et al. (2026) Job-shop scheduling with resource flexibility: a systematic review from traditional to AI-integrated approaches. Journal of Industrial Information Integration, 51. https://doi.org/10.1016/j.jii.2026.101093 [32] Vaz, L.V., Gonçalves, M.C., Dias, I.C.P. and Nara, E.O.B. (2024) Application of a production planning model based on linear programming and machine learning techniques. Journal of Engineering and Technology for Industrial Applications, 10(45). https://doi.org/10.5935/jetia.v10i45.920 [33] Wahid, A., Breslin, J.G. and Intizar, M.A. (2022) Prediction of machine failure in Industry 4.0: a hybrid CNN-LSTM framework. Applied Sciences, 12(9). https://doi.org/10.3390/app12094221 [34] Wheeler, M.R., Everett, B. and Prybutok, V. (2026) Navigating the environmental paradox of AI: a decision framework for clean technology practitioners. Clean Technologies, 8(2). https://doi.org/10.3390/cleantechnol8020051 [35] Yang, H., Yuan, J., Li, C., Zhao, G., Sun, Z., Yao, Q. et al. (2022) BrainIoT: brain-like productive services provisioning with federated learning in industrial IoT. IEEE Internet of Things Journal, 9(3). https://doi.org/10.1109/JIOT.2021.3089334 [36] Zhang, C., Wu, Y., Ma, Y., Song, W., Le, Z., Cao, Z. et al. (2023) A review on learning to solve combinatorial optimisation problems in manufacturing. IET Collaborative Intelligent Manufacturing, 5(1). https://doi.org/10.1049/cim2.12072 [37] Zhang, R., Song, S. and Wu, C. (2020) Robust scheduling of hot rolling production by local search enhanced ant colony optimization algorithm. IEEE Transactions on Industrial Informatics, 16(4). https://doi.org/10.1109/TII.2019.2944247 [38] Zhang, Y., Zhu, H., Tang, D., Zhou, T. and Gui, Y. (2022) Dynamic job shop scheduling based on deep reinforcement learning for multi-agent manufacturing systems. Robotics and Computer-Integrated Manufacturing, 78. https://doi.org/10.1016/j.rcim.2022.102412

Copyright

Copyright © 2026 Dr. Richa Kalpesh Saxena, Shrawani Behere, Shiv Karbhari, Shri Choksi, Shubham Parekh, Smit Shah. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.