The term Industry 4.0 refers to the fourth industrial revolution which brings together rapidly evolving technologies such as the internet of things(IoT),artificial intelligence(AI),robotics, and advanced computing to dramatically change the landscape of manufacturing.
Industry 1.0
Industry 1.0 saw the manual processing of botanical, mineral, and animal derived materials transition from simple hand-operated tools to commercial-scale machinery able to crush, mill, blend ,and press larger quantities of medicines (Anderson,2005). In the 19th century, larger-scale pro- duction of drugs utilizing non-electrical power-driven machinery emerged from two sources– individual pharmacies or the dye and chemicals industry (SonnedeckerandUrdang,1976;DaemmrichandBowden,2005).
Industry 2.0
The second industrial revolution was enabled by electricity and early electronic machines and assembly lines with pre-set controls that incorporated basic automation and process controls which provided manufacturers the ability to set basic process parameters. In the pharmaceutical manufacturing industry, this manifested as electronic machine-based crushing, milling, blending and tablet pressing allowing for larger-scale production and– importantly– more monitoring of processes and quality. Industry 2.0 developments led directly to machines such as modern tablet presses that can reliably produce over one million tablets per hour (Berry and Nash, 2003).
Industry 3.0
The third industrial revolution was enabled by the development and availability of computers and communication technologies, such as networked computing, the internet, and wireless communications. These technologies enable a higher degree of automation of processes and equipment, which in pharmaceutical manufacturing enabled concepts such as continuous manufacturing and active control. Human- computer interfaces aided in developing more sophisticated control strategies and higher product and process quality. Remote sensing and monitoring reduced the need for human operators on the manufacturing floor and facilitated better tracking of parameters and metrics associated with production.
Some industries are now well into Industry 3.0, but in many ways the pharmaceutical industry is still very much transitioning into it. For example, continuous manufacturing is a technology that sends materials produced during each process step directly and continuously to the next step for further processing; it has been widely adopted in other industries. For various reasons, the pharmaceutical industry has been slower to adopt continuous manufacturing (Lee et al., 2015). As a corollary, the pharma industry has also not yet achieved consistent six sigma manufacturing capability (i.e., <3.4 errors per million opportunities) which is common in other industries (Yu and Kopcha, 2017).
The third industrial revolution brought pharmaceutical manufacturing advanced process analytical technology (PAT),which aims to provide process and product quality data in near real time.
In-dustry3.0alsoadvancedmodel-basedorQualitybyDesign (QbD) processes, which aim to control target product quality profiles within a defined set of quality parameters.
Industry 4.0
Industry 3.0 saw rapid advancements of individual operations and tools, Industry
4.0promises advancements of entire manufacturing systems and infrastructures.
In such an environment, performance data can be analyzed by algorithms and used for critical real-time business and operational decisions that directly impact production outputs.
The archetypal feature of an Industry 4.0 environment is the integration of connectivity, artificial intelligence (AI), and robotics to enable systems that operate with little to no human involvement (Leurent anddeBoer,2018).Integratedautonomousandroboticsystemsfusereal-timeandonlinedatawithindustrialproductionprocessesandartifi-cial intelligence in order to optimize manufacturing and enterprise-wide management (Moore,2018).Multiple data sources can integrate to connect both external and internal information. For example, in pharmaceutical manufacturing, external information including variables such as patient experience, market demand, supplier inventories, and public health emergencies - could fuse with internal information such as energy and resource management, modeling and simulation outcomes, and laboratory data. Integrating internal and external data sources enables unprecedented real-time responsiveness, monitoring, control, and prediction(Fig.2).
Fig. 1. The stages of data transformation on the path to realizing Industry 4.0. In these stages, data are transformed from raw signals captured from a system to full digital maturity. Data are initially collected from a manufacturing process, then organized by data digitization and analysis as Big Data into information, then synthesized into knowledge by the meaning discerned via artificial intelligence, and finally to actionable wisdom attained through the combined insights of digital maturity.
Fig. 2. A cyber-physical system (CPS) for pharmaceutical manufacturing in Industry 4.0. Key parts of a CPS include the public-cloud, private-cloud, and manufacturing floor. The public cloud contains application services for external customers. The private cloud deals with information for higher layer features such as remote monitoring systems, production, energy management, laboratory information, control service, and modeling and simulation. The public and private cloud digitally reflects the status of the physical system, thereby enabling real-time optimizations and predictions. The manufacturing floor consists of equipment, PATs instrumentation, and real-time release testing (RTRt). PAT provides control of the manufacturing process and RTRt ensures product quality based on the information collected during the manufacturing process. The operations of the process (e.g., feeding, wet granulation, fluid bed drying, milling, blending, compression, and tablet coating) connected to the local network and the cloud through the internet.
Fig. 3. The enabling technologies of an Industry 4.0 smart factory. Data from a manufacturing process are captured and stored via two key technologies: data storage technology, such as the cloud, and data capture technology, such as advanced sensors used in an operation. Data storage technology enables the long-term storage of digitized data captured from advanced sensors.This data-rich environment enables Big Data and simulations, artificial intelligence and adaptive control, digital twins, and cyber-physical systems like the internet of things. These combined technologies enable intelligent, precision, real-time, collaborative robotics and augmented or virtual reality technologies to run and manipulate manufacturing. The entirety of the smart factory is enabled by a wireless internet network and appropriate cybersecurity.
Digitization and digital maturity
A key to implementing Industry 4.0 is the digitization of multiple complex pieces of the pharmaceutical value chain with embedded cybersecurity. A critical concept in developing the so-called “smart factory” is the industrial internet of things (IoT), which is a type of cyber-physical system comprising interconnected computing devices,
sensors, instruments, and equipment integrated online into a cohesive network (IEEE, 2021). The IoT requires data digitization, which is the transformation of previously manually captured data to digital device- captured data. In pharmaceutical manufacturing this may include sup- ply chain-related information such as raw materials variability and global tracking of materials across facilities (Sandle, 2019; Marcus Ehrhardt PB, 2016), manufacturing floor-related information such as operation procedures and operator work instructions (Jovanis, 2019), monitoring real-time operations by video (Marcus Ehrhardt PB, 2016), video-based training and centralizing quality event data for improved decision making (Jovanis, 2019). Full digital maturity, the process of gaining wisdom from these digitized data, is necessary to transform reactive operations into a fully integrated and digital ecosystem capable of proactive and predictive decision making (Grossman, 2018). This integration enables real-time connectedness both within a manufacturing facility(e.g., machine learning across unit operations) as
well as outside the facility, as products “talk” back to their manufacturers using technologies that track environmental conditions, quality attributes, use, and performance of products (PwC, 2015, 2016). Together with AI algorithms focusing on machine learning and adaptive control (described below), the IoT would be disruptive in pharmaceutical manufacturing and product development (Biophorum/BPOG, 2017) (Fig. 3).
Industry 4.0 may well shift the key problem for pharmaceutical manufacturers from controlling processes to enabling human understanding of the operations. To facilitate visualization and human understanding of digitized manufacturing operations, each segment of the value chain could be divided into digital architectures; i.e., hardware, and software infrastructures that support data capture, storage, and analysis. Different types of architectures have been proposed to address this problem including digital ecosystems or digital compasses (i.e.,
different modes of organizing how data is collected, stored, and analyzed) that highlight ‘levers’ mapping to key value drivers (Baur and Wee, 2015; Hartmann et al., 2015; PwC, 2021). In order to develop cost- effective intelligent systems that merge online data with production systems and customer demand, it will be necessary to further develop
powerful computing architectures (i.e., sets of rules and methods that describe functionality, organization, and implementation) and improve high-speed communications, at lower costs (Lydon, 2017). While the specific tools developed to optimize key value drivers may vary depending on the core competencies and business models of pharmaceutical firms, the development of more integrated systems will be consistent. It is the digital integration into an IoT that could produce disruptive pharmaceutical applications such as real-time, on-demand, small-scale production systems, truly personalized dosage forms, and revolutionary biosensor diagnostics (Tracy, 2017; Wilson, 2018; Woods, 2017; Banks, 2015).
Artificial intelligence
AI involves the integration of digital data and computational analysis for the purposeof making decisions normally made by humans (McCarthy and Thomason, 1989). Tasks that rely on computer-based intelligence may involve reasoning, problem solving, learning, and decision-making among others. The application of AI in pharmaceutical manufacturing has already begun with examples including the use of machine vision technology (Forcinio, 2019; Veillon, 2020; Yadav, 2020) to replace human visual inspection of packaging, caps and vials; predictive equipment maintenance to reduce disturbances, risks, and pro- duction downtime(Otto, 2019; Markarian, 2020); and automated quality control enabling seamless scheduling of analytical testing(Han et al., 2019), continuous process quality assurance (Chapman, 2019), and enhanced data integrity (Powers, 2019).
AI includes a spectrum of sub-disciplines which take varied approaches to designing computer intelligence depending on the desired features and tasks to be performed. Such approaches involve handling large and disparate datasets with specific algorithms. Within the field of AI, and due to the advancements in available technology and software programming, machine learning (ML) and artificial neural networks (ANN) have emerged as two of the more advanced methods for pre- diction and risk management. In the hierarchical relationships of AI, ML is a sub-discipline of AI, and ANNs are a sub-discipline of ML.
ML primarily involves the ability of computers to learn a task by monitoring data and using statistical tools in order to derive some general knowledge from these data (via the development of mathematical relationships) without external input or prompt (McCarthy et al., 2004). It is worthwhile to note that ML algorithms can fall into one of three categories depending on how input data are utilized: supervised learning, unsupervised learning, and reinforcement learning. Supervised learning can include methods such as ANN or multivariate regression and classification analysis, which learn from and connect input data and outcomes. Supervised learning methods are commonly associated with process design and controls. Unsupervised learning draws inferences from input data without using outcomes to learn. Unsupervised learning approaches, such as dimension reduction or cluster analysis, are useful in identifying trends and anomalies associated with an operation. Reinforcement learning correlates actions with delayed outcomes so that decisions are associated with desired outcomes in the future. Reinforcement learning can be used where complex dynamics are involved; for example, plant operations, or logistics. While each of these ML approaches has the potential to enhance pharmaceutical manufacturing operations, supervised learning approaches are typically viewed to have less risk and uncertainty and have thus far gained the most traction.
Supervised learning ML approaches such as ANN have seen steady progress in advanced manufacturing applications (Peres et al., 2016; Arinez et al., 2020). ANNs are modeled after the connectivity between the neurons and synapses of the human brain which utilize data-driven algorithms to determine a mathematical relationship between input and output variables. The design and structure of the ANN is such that individual nodes in one layer are connected via weighted connections to individual nodes in subsequent layers. ANN models can be developed and applied independently or, as is often the case, utilized in conjunction with other modeling techniques. ANN has been used for prediction and control in pharmaceutical development (Ekins, 2016; Korteby et al., 2016) and recently to perform risk-based analysis of biomanufacturing processes (Shirazian et al., 2017; von Stosch et al., 2016a, 2016b), develop control schemes and perform fault detection for complex dy- namic processes (Montague and Morris, 1994; Shimizu et al., 1998; Stanke and Hitzmann, 2013; Takahashi et al., 2015), and to predict outcomes for therapeutic drug pharmacokinetics and pharmacodynamics (Atobe et al., 2015; Lin et al., 2015; Pavani et al., 2016; Yama- mura, 2003). A significant advantage of ANN models is their utility in pattern recognition within a dataset – even with noisy or complex data with missing data points.
Computer vision quality control, digital twins, predictive maintenance, real-time augmented reality, and collaborative robots are tools better enabled by AI (Fig.3). AI should generally improve and optimize manufacturing processes while also reducing human intervention in the production of pharmaceuticals. Computer vision-based quality control uses images (for example, images of packaging, labels or glass vials) that are analyzed by software to detect deviations and to ensure images match the requirements of a given quality attribute of a product.
Collaborative robots (i.e., cobots: groups of robots programmed to work together) act in collaboration through one or more integrated software programs in order to achieve a desired outcome through a series of steps such as packing, moving and sealing a box or taking a sample of material from process machinery, moving it to a different location, analyzing it, and sending information back to the process machinery controls (Fig. 3).The use of augmented reality may be useful in the areas of customer experience, discovery and research, maintenance, quality assurance, safety, packaging and training
A digital twin is a digital replica of a physical process such as an operation, machine or activity used to better understand, evaluate, predict, and optimize its performance (Fig. 3). Digital twins can be based on empirical data (data-driven models) or integrate both empirical and mechanistic simulations to provide high resolution models together with real-time or near real-time data from which to assess process performance. Such models outperform traditional process models both in terms of resolution and real-time feedback. For example, some companies outside pharma have employed digital twins in smart factories (Wilson, 2020; GE, 2020) and inside pharma in smart processes (InSilico, 2020). Digital twins enable humans to better understand how deviations or disruptions may impact performance, and how related risks can be mitigated.
Full automation
Complete process automation includes the capture of all process performance data via integration of cloud-connected PAT technology, followed by the analysis of that data into organized information, the application of AI based algorithms to convert that information into knowledge, and finally the use of that knowledge to gain insight about the process and to enhance process control (Fig. 1). A useable IoT re- quires the capacity for individual units to connect to the cloud in order to send and receive data (Fig. 3). Within the current landscape of pharmaceutical manufacturing, process controls are typically segregated from process performance. This leads to an inherent delay in applying modifications to control systems, for example in a situation where process performance trends out of specification. This challenge may be mitigated by the application of AI algorithms employing ML or ANN approaches using process data to detect and predict when measured parameters are trending out of specification and making changes before they do so. ANN models can be utilized for such adaptive control of processes due to their ability to learn, predict, and forecast process states based on current and historical data. The application of these concepts will enable the real-time capture of process performance parameters and trends in the data, which can further be applied to predict product quality attributes further down the process pipeline.
This adaptive control strategy enables the additional opportunity for real-time process optimization via the application of digital twins(Fig. 3).
Conclusion
Industry 4.0 technologies have the potential to transform pharmaceutical manufacturing and logistics platforms through digitization, autonomous systems, robotics, and computing advancements. In particular, the pharmaceutical supply chain, production processes, distribution, and inventory frameworks could see significant improvements. The smart factory of the future will take on autonomous features enabling more production flexibility and agility. The path to full adoption of Industry 4.0 will require advancements and innovations addressing multiple data, computing, and automation risks and challenges. Both industry and regulators are developing competencies in preparation for smart manufacturing systems: modeling and simulations, sensor systems, data management, data analytics, computational and control engineering approaches needed to support autonomous systems, artificial intelligence, and computing infrastructures. Enterprise-level systems such as quality management and training may need to be re-envisioned. We are beginning to identify known or latent risks associated with these new approaches, recognize dissonance with existing approaches to regulatory compliance, and ultimately develop regulatory frameworks that support Industry 4.0.
References
1. Anderson, S., 2005. Making Medicines: A Brief History of Pharmacyand Pharmaceuticals. Pharmaceutical Press.
2. Arinez, J.,Chang, Q., Gao, R., Xu, C., Zhang, J., 2020. Artificial Intelligence in Advanced Manufacturing: Current Status and Future Outlook. J. Manuf. Sci. Eng. 142 (11), 16. Atobe, T., Mori, M., Yamashita, F., Hashida, M., Kouzuki, H., 2015. Artificial neural
network analysis for predictinghuman percutaneous absorption taking account of vehicle properties. J. Toxicolog. Sci. 40 (2), 277–294.
3. Balasegaram, M., Kolb, P., McKew, J., et al., 2017. An open source pharma roadmap.
PLoS Med. 14 (4), e1002276.
4. Banks, L., 2015. 10 Disruptive Technologies that will Transform Pharma. Pharmaphorum. Baur, C., Wee, D., 2015. Manufacturing’s Next Act. McKinsey & Company.
5. Berry, I.R., Nash, R.A., 2003. Pharmaceutical process validation. In: Nash, Robert A., Wachter, Alfred H. (Eds.).
6. Biophorum/BPOG, 2017. Automated Facility https://www.biophorum.com/wp-content/ uploads/2017/07/Automated-Facility.pdf. Accessed.
9. Ekins, S., 2016. The Next Era: Deep Learning in Pharmaceutical Research. Pharm. Res. 33 (11), 2594–2603.
10. Ezell, S.A., 2016. Policymaker’s Guide to Smart Manufacturing. Information Technology Innovation Foundation.
11. Fassbender, M., 2017. More Than Entertainment: Augmented Reality in the Pharmaceutical Industry.