Introduction
Related work
Smart manufacturing
-
Knowledge-embedded facilities—embedding traditional facilities with knowledge and intelligence with the aim of enabling ‘smart’ operations across the factory.
-
Predictive and preventive operations—transforming operations from reactive and responsive, to those that are predictive and preventative.
-
Performance-based operations—promoting performance over compliance, with an emphasis on minimising energy and material usage, while maximising sustainability, health and safety, and economic competitiveness.
-
Distributed intelligence—replacing vertical and localised decision-making with collective intelligence, where decisions are made to benefit the entire organisation.
-
Multi-disciplinary workforce—removing the boundaries of vertical factory operations and cultivating a culture of an interdisciplinary workforce.
-
Next-generation IT departments—retraining personnel to handle data-intensive, real-time and internet-aware infrastructures and technologies, such as real-time IoT sensors and big data technologies.
Groups and initiatives
Impact and benefits
Roadmap for smart manufacturing
-
Phase 1—data integration and contextualisation. Initially, facilities evaluate what data is available in different areas of the factory (e.g. sensors, controllers, databases, etc.) to form a global and contextualised view of data in the facility. Similarly to business intelligence and data warehouse projects, the process of integrating data can be a time-consuming and complex task. However, in manufacturing environments this complexity can be exacerbated due to the wide-range of industrial devices and protocols that can exist. The benefits of data integration and contextualisation alone may have a positive impact on operating costs, health and safety, and environmental factors.
-
Phase 2—simulation, modelling, and analytics. Once data has been integrated and contextualised it can be processed and synthesised to create manufacturing intelligence that can inform decision-making. While data integration and contextualisation increases data visibility throughout the factory, data processing can be used to formulate actions that positively affect operations, which eclipse the benefits of fundamental data management alone. The potential benefits derived from advanced data processing include flexible manufacturing, optimal production rates and enhanced product customisation.
-
Phase 3—process and product innovation. As manufacturing intelligence is accumulated from data processing, new insights will begin to emerge from repositories of collective intelligence. In turn, these insights can inspire major innovations in processes and production. The benefits derived from such insights will be capable of causing significant market disruption that result in game-changing economics (e.g. 90 % reduction in retail price of laptop).
Impediments and challenges
-
Historical investment in IT and automation Over the last 40 years many facilities invested in IT systems and automation networks. Therefore, facilities may be reluctant to replace technologies that received significant investment, and continue to function at an appropriate level.
-
Regulatory and quality constraints In certain industries, such as pharmaceutical and medical devices, internal or external regulatory and/or quality standards may restrict the adoption of new technology. While in some scenarios this impediment can be circumvented through policy alteration, the exhaustive procedures associated with such a process may negate the enthusiasm for technology replacement.
-
Dependency on proprietary systems or protocols Although many manufacturing information systems and automation network standards exist, such as ISA95 for system interoperability and OPC for device communication, they are not always adopted. Therefore, when facilities are locked-in to proprietary and closed technologies, rather than those built on open standards, technology adoption (i.e. smart technologies) may be restricted to the offerings of their vendor.
-
Weak vision and commitment Transitioning to smart manufacturing requires strong leadership and a shared vision of its short and long-term benefits. Facilities that do not have a clear understanding of how smart manufacturing can impact operations, may not have an appetite to replace equipment and technologies that are currently operating as intended.
-
High risk and disruption There is a high-risk associated with new technology and system implementation. These type of projects may negatively impact operations while personnel are achieving competency or fail to operate as originally intended. Therefore, the desire to undertake large-scale IT projects for smart manufacturing adoption may remain weak until such time lost opportunities affect competitiveness.
-
Skills and technology awareness IT, automation and facilities departments are entrenched in legacy computing, automation and networking methods that have been in existence for decades. However, smart manufacturing adoption requires a significant shift from these approaches. Therefore, unless these new technologies and methods are embraced, smart manufacturing adoption may be severely impeded.
-
Multi-disciplinary workforce A strong multi-disciplinary workforce is an important aspect of smart manufacturing, where key decision-makers may need knowledge from multiple disciplines, such as engineering, computing, analytics, design, planning, automation, and production [15, 18]. Multi-disciplinary personnel may be particularly important for demand-driven supply chains, large-scale data analysis, system interoperability, and cyber physical systems [11].
Industrial equipment maintenance
Maintenance strategies
Strategy | Intent | Benefits | Weaknesses | Suitability |
---|---|---|---|---|
Reactive | Only undertake maintenance when a complete equipment failure occurs | No upfront planning or scheduling | Unpredictable equipment availability, shorter equipment lifetime, increased energy costs, and potentially lower production yield due to partial malfunctions | Suitable for non-essential equipment, or in situations where the cost of maintaining equipment is greater than the cost of failure |
Corrective | Identify and address individual/minor faults when they occur to avoid a complete equipment failure | Manages risk of complete failure, provides visibility of equipment health, and can increase lifetime of equipment through timely maintenance activities | Investment in diagnostic technologies, as well as the labour cost associated with monitoring and managing faults | Suitable for broad classes of equipment maintenance, but may not be suitable for mission critical equipment, where a complete failure must be mitigated at all costs |
Preventative | Perform regular maintenance to avoid either partial or complete equipment failures. Preventative maintenance can be undertaken at time intervals (e.g. change component X every 4 weeks regardless of its state), or when a particular condition has been met (e.g. heating element begins to take X minutes to reach its target temperature) | Promotes confidence in machine availability by mitigating equipment failure using pre-determined maintenance intervals/conditions | Prematurely replacing components and carrying out maintenance activities may come at the expense of high maintenance costs, or at least costs that are sub-optimal | Suitable for scenarios where every attempt must be made to ensure that mission critical equipment is available and operating correctly at all times, but this is typically done at the expense of resource efficiency and cost |
Predictive | Predict an issue before it occurs and be capable of estimating the remaining useful life (RUL) of the equipment and/or its internal components | Optimises resources and reduces costs by predicting the lifetime of components to avoid premature replacement and circumvent redundant maintenance activities | Given prediction is probabilistic rather than deterministic, there is potential for false positives that could lead to unnecessary maintenance actions | Suitable for scenarios where the operation, cost and output derived from equipment must be optimised, but occurrences of machine availability can be tolerated |
Issue identification techniques
Technique | Strategies | Description | Implementation | Weakness |
---|---|---|---|---|
Basic intelligence and reporting | Corrective | Reporting is used to manually assess if a particular parameter is outside an expected operating boundary. If so, further investigation of the potential issue can be undertaken | Existing industrial information systems, such as Manufacturing Execution Systems (MES) or Building Management Systems (BMS), which are used in day-to-day operations, can be used to generate reports | Largely manual process, with static and ad hoc issue identification. Also dependent on the ability of the expert analysing the report to observe the anomaly, which may be somewhat subjective and easy to overlook due to human error |
Fault detection and diagnosis (FDD) | Corrective | FDD consists of a set of encoded fault logic (e.g. IF/THEN rules), to identify potential issues based on a set of input data | FDD capabilities are embedded in some industrial information systems, but also exist as standalone systems and tools that can be used to monitor specific types of equipment | Logic employed is typically specific to equipment, and detection means that the issue is already present and may be impacting operations in some way |
Condition-based monitoring (CM) | Corrective Preventative | CM focuses on monitoring a particular measurement, or set of measurements, to determine if an issue has, or is likely to occur. The condition is fired when the monitored parameter(s) are outside a predefined range | CM is available in many modern industrial information systems, and can be viewed as an extension to reporting and monitoring modules, with a condition/trigger that automatically highlights issues | The condition is specific to equipment and/or its components. Therefore, performance and accuracy is dependent on the appropriate parameter(s) being chosen, and condition values set |
Predictive maintenance (PM) | Preventative Predictive | PM employs statistical learning techniques to anticipate the occurrence of an issue, and/or estimate the RUL of equipment and components | Predictive methods in current industrial information systems are limited, and therefore, PM it is common to see implementations as standalone systems or tools | To develop an accurate tool, an appropriate amount of high-quality data must be available to inform the statistical learning model |
Prognostics and health management (PHM) | Corrective Preventative Predictive | PHM uses a holistic approach to issue identification, and comprises FDD, CM, and PM, to highlight issues at different stages so that optimal equipment health is maintained | Given multiple techniques are used in PHM; it is typically implemented as a dedicated system. In some cases, where interoperability exists, PHM systems may leverage the FDD, CM or PM capability of an existing system | Implementing multiple techniques represents challenges – e.g. when should a particular technique be used. This arguably makes PHM more complex than any single technique |
Industrial information and data systems
Classification | Data source | Data type | Common interfaces | End user | Latency |
---|---|---|---|---|---|
Database | Building management system (BMS) | Energy and environmental | ODBC OLEDB JDBC etc. | Energy and facilities | Batch |
Database | Monitoring and targeting (M&T) | Energy | ODBC OLEDB JDBC etc. | Energy and facilities | Batch |
Database | Manufacturing execution system (MES) | Production and automation | ODBC OLEDB JDBC etc. | Automation, production and quality | Real-time and batch |
Device | Programmable logic controller (PLC) | Production, energy and environmental | OPC MT Connect BACnet Modbus LonWorks | Automation and building services | Real-time |
Device | Gateways | Multiple | HTTP OPC Modbus I/O (i.e. CSV) | Multiple | Real-time and batch |
Device | Smart devices (i.e. IoT) | Multiple | MQTT COAP HTTP | Multiple | Real-time |
Research methodology
Research scope
Type of applications
Regulation and compliance
Time-series data
Immutable data
Requirements analysis
Industry partner
-
Automation—the automation team comprised personnel with knowledge of control, production, energy and information technology. Liaising with the automation team provided a better understanding of the technologies, infrastructure and procedures relating to production, scheduling and maintenance.
-
Energy—the energy team comprised personnel with specialist knowledge of engineering, energy and environment. The energy team demonstrated current systems for measuring the energy consumption of machinery, and illustrated how malfunctioning equipment can negatively impact operating cost.
-
Big data and smart manufacturing—no single team was responsible for big data and smart manufacturing, which is understandable given the contemporary and interdisciplinary of both fields. However, several teams were considering applications of big data in manufacturing, while other teams were investigating strategies for smart manufacturing. These teams showed a positive attitude towards smart manufacturing approaches, with plans to integrate IoT, big data systems and data-driven analytics being a recurring theme.
Research questions
RQ1: How can industrial big data analytics applications be enabled and integrated in existing large-scale manufacturing facilities?
RQ2: What type of data architecture is needed to support the implementation of industrial big data analytics focused on equipment maintenance?
Results
RQ1: Requirements and challenges
Legacy integration
Cross-network communication
Fault tolerance
Extensibility
Scalability
Openness and accessibility
RQ2: Data pipeline architecture
Stage 1: Site manager
Stage 2: Ingestion process
Stage 3: Message queue
Stage 4: Subscription service
Stage 5: Data processing
Stage 6: Data access
-
Data—refers to a particular data set that was processed (e.g. energy).
-
Object—an identifier within the data set (e.g. equipment identifier).
-
Year—the year that is relevant to the query.
-
Month—the month that is relevant to the query.
-
Day—the day that is relevant to the query.
Discussion
Part 1 of simulation: data ingestion
Requirement | Discussion |
---|---|
Legacy integration | The simulation illustrates how legacy and smart devices can coexist in the industrial big data pipeline, with legacy integration realised using the ingestion engine to abstract and encapsulate legacy devices and instrumentation relating to the RAT measurement. Both measurements are pushed to the cloud for industrial analytics applications to consume without being aware of their origin (i.e. legacy or smart device) |
Cross-network communication | Ingestion engines can operate across different networks due to lack of local dependencies and cloud data processing. The simulation shows that an ingestion engine can be deployed on a network with an active internet connection, access to the site manager, and reachable data sources. Considering the flow of data depicted in the simulation, the network location of the ingestion engine and smart sensor are irrelevant as both measurements reach the same endpoint (i.e. cloud platform) |
Fault tolerance | Reliability and resilience can be instilled by adding more ingestion engines to the process and monitoring the status of each. As illustrated by the simulation, the site manager controls the ingestion process by sending data collection instructions to the ingestion engine. In scenarios where an ingestion engine fails to complete its assigned task, the instructions can be reassigned to another ingestion engine. This promotes fault tolerance in the ingestion process by removing a single point of failure, while allowing for different levels of resilience |
Extensibility | It is feasible that the data pipeline may need to be extended to support additional data sources. As ingestion engines encapsulate data integration logic they are the logical point of extension. Given the existence of an abstract function that expects data collection instructions from the site manager as input, and outputs JSON encoded measurements, additional data sources could be facilitated by implementing new instances of the function |
Scalability | The ingestion process can be scaled horizontally by deploying additional ingestion engines across machines and networks. This provides the data pipeline with a greater work capacity given that more ingestion jobs can run in parallel. The ingestion engine characteristics that make this possible have already been addressed in discussions regarding cross-network communication and fault tolerance |
Openness and accessibility | The simulation shows how open standards, such as HTTP and JSON, can be used to facilitate communication and data exchange amongst distributed components in the data pipeline. This is exemplified by communication between the site manager and ingestion engine to relay data collection instructions, as well as the transmission of measurements to the cloud from the ingestion engine and smart sensor |
Part 2 of simulation: data processing
Requirement | Discussion |
---|---|
Legacy integration | The simulation shows JSON encoded RAT and SPT measurements in the message queue. It is apparent that legacy integration was successful given the RAT measurement originated from a legacy source, but now resides in the same format as SPT |
Cross-network communication | The unified message queue shown in the simulation illustrates how messages from across devices and networks are assimilated after a successful ingestion process |
Fault tolerance | Cloud computing provides a fault tolerant environment for data processing due to its ability to scale resources based on demand. The data pipeline simulation depicts a high-throughput cloud infrastructure with data processing and message queue components. It is implied that these components reside in a highly distributed cloud service that provides fault tolerance across multiple compute nodes. Furthermore, the simulation clearly shows how the message queue decouples the data ingestion from data processing. This promotes additional fault tolerance by protecting the ingestion process from faults that originate from data processing components |
Extensibility | The simulation illustrates the aggregation and contextualisation component that is responsible for preparing data for analysis. This is one example of processing that may be required for time-series data. As new processing needs arise additional components can subscribe to the message queue subscription service and begin processing in parallel with other components. This extension is facilitated by the decoupling of processing components and message queue using the subscription service |
Scalability | The data processing simulation inherits its scalability from its cloud infrastructure. Many of the benefits of cloud computing have already been discussed with regard to fault tolerance. These same load balancing features facilitate scalable data processing in the pipeline by scaling compute resources based on demand (i.e. amount of processing required) |
Openness and accessibility | The simulation shows data stored in a contextualised cloud repository after completion of the aggregation process. This repository uses a naming convention to identify a dataset (i.e. AHU1), object (i.e. RAT), and chronological association for accessing time-series data. To promote openness and accessibility this data is accessed using standard HTTP requests. Furthermore, the ability of the data pipeline to support additional data formats, standards and representations has been addressed in discussions regarding extensibility |
Part 3 of simulation: industrial analytics
Requirement | Discussion |
---|---|
Legacy integration | The previous parts of the simulation ensured that legacy data integration was achieved. Data access illustrated in this simulation shows both applications are able to access measurements regardless of whether they originated from legacy or smart devices |
Cross-network communication | Similarly, previous parts of the simulation have realised cross-network communication. Data ingestion undertaken across all networks is unified by message queues before being exposed to applications for reading |
Fault tolerance | The inherent qualities of cloud computing are used to provide fault tolerance in the data access part of the simulation. For example, file storage and delivery services in the cloud (e.g. Amazon S3) can provide a distributed, low latency and fault tolerant platform for serving time-series data |
Extensibility | Extensible data access in the pipeline is necessary to service industrial analytics applications. For example, an application may require a certain data format or standard to be presented. In this scenario the new format can be generated by a processing component (part 2 of simulation) and pushed to the cloud repository for industrial applications to access |
Scalability | Similarly to fault tolerance, the scalability of the data access part of the simulation is dependent on the cloud service on which it resides. The ability of cloud-based file delivery services to scale horizontally across multiple compute nodes and data centres provides a highly scalable infrastructure for serving time-series data |
Openness and accessibility | The simulation illustrates how data access can be achieved from the data pipeline with context encoded URL’s over HTTP. Furthermore, no proprietary or commercial technologies or drivers are required to consume the time-series data from the cloud. Therefore, there are no obvious technology barriers preventing users, applications and systems from accessing the data |