Predictive maintenance and asset availability optimization

Improve asset maintenance with real-time insights

Predictive maintenance is a key component of Industry 4.0. Poor maintenance strategies can substantially affect the operational efficiency and profitability of industrial manufacturers. To be competitive, companies in asset-intensive industries need to minimize unplanned downtime and optimize maintenance costs.

• 82% of companies have had unplanned downtime in the past three years, costing as much as US$260,000 per hour, with outages lasting an average of four hours.
• For asset-intensive organizations, the maturity of their maintenance practices is a key determinant of their ability to operate reliably, without interruption, and profitably. Improving maintenance practices, processes, and systems can deliver a large return on investment.
• Organizations can use predictive analytics to predict asset failure and reliable lifespan and generate actionable insights in real time.

For the manufacturing industry, using data to enable and improve predictive maintenance is particularly relevant as the use case can be applied to any kind of manufacturing production system, such as computerized numerical control (CNC) infrastructure, supply chain and warehouse systems, logistics and test systems, and so on.

While a wide variety of data sources may be used depending on the specific application, the keys to shifting from reactive to predictive maintenance are Internet of Things (IoT) data streams or machine-to-machine (M2M) messages sent and received through an MQTT (the IoT messaging standard) broker or provided by historians from operational intelligence systems. These are the sources of the raw data required to assess whether maintenance operations are needed; however, data from other sources is necessary to establish a proper predictive maintenance system. For instance, maintenance management systems contain information about the pieces of equipment themselves, such as maintenance reports. Other sources of data include supervisory control and data acquisition (SCADA) systems, a special repository containing media files (such as pictures and video streams), maintenance manuals, and weather forecasts. The variety of data that can be used in predictive maintenance is vast.

Reduce costs and improve efficiency by fine-tuning predictive maintenance

The architecture presented here demonstrates how recommended Oracle components can be combined to build a full analytics architecture that covers the entire data analytics lifecycle, from discovery through to action and measurement, and delivers the wide range of business benefits described above.

This image shows how Oracle Data Platform for manufacturing can be used to support predictive maintenance and asset availability optimization. The platform includes the following five pillars:

• Data Sources, Discovery
• Ingest, Transform
• Persist, Curate, Create
• Analyze, Learn, Predict
• Measure, Act

The Data Sources, Discovery pillar includes two categories of data.

Business record data comprises data from MES, WHM, CMM (maintenance and asset management), IoT, SCADA systems, and historian and operator entry (including faults, quality, and observations).

Technical input data includes IIoT, images, email, videos, paper documentation (OCR), and discrete events (such as an emergency stop of the production line).

The Ingest, Transform pillar comprises four capabilities.

Batch ingestion uses OCI Data Integration, Oracle Data Integrator, and DB tools.

Bulk transfer uses OCI FastConnect, OCI Data Transfer, MFT, and OCI CLI.

Change data capture uses OCI GoldenGate.

Streaming ingest uses Kafka Connect.

All four capabilities connect unidirectionally into the serving data store, transactional data store, and cloud storage within the Persist, Curate, Create pillar.

Additionally, streaming ingest is connected to stream processing within the Analyze, Learn, Predict pillar.

The Persist, Curate, Create pillar comprises five capabilities.

The serving data store uses Autonomous Data Warehouse and Exadata Cloud Service.

The transactional data store uses ATP, MySQL, Oracle NoSQL, and Exadata Cloud Service.

Cloud storage uses OCI Object Storage.

Batch processing uses OCI Data Flow.

Governance uses OCI Data Catalog.

These capabilities are connected within the pillar. Cloud storage is unidirectionally connected to the serving data store and the transactional data store; it is also bidirectionally connected to batch processing.

Two capabilities connect into the Analyze, Learn, Predict pillar. The serving data store connects to both the analytics and visualization capability and also to the data products, APIs capability. Cloud storage connects to the machine learning capability.

The Analyze, Learn, Predict pillar comprises five capabilities.

Analytics and visualization uses Oracle Analytics Cloud, GraphStudio, and ISVs.

Data products, APIs uses OCI API Gateway and OCI Functions.

Machine learning uses OCI Data Science and Oracle Machine Learning.

AI services uses OCI Anomaly Detection, OCI Forecasting, OCI Language, and OCI Vision.

Streaming processing uses GoldenGate Stream Analytics and stream analytics from third parties.

The Measure, Act pillar captures how the data analysis may be used: by people and partners and applications and models, and specifically to update AI service models.

People and partners comprise Condition Monitoring and Sensor Data Analysis, Failure Mode and Effects Analysis (FMEA).

Applications comprises Asset Performance Management (APM), Root Cause Analysis, Reliability Centered Maintenance (RCM).

Models comprises Updated AI Service Model, Predictive Analytics and Machine Learning Models

The three central pillars—Ingest, Transform; Persist, Curate, Create; and Analyze, Learn, Predict—are supported by infrastructure, network, security, and IAM.

Connect, ingest, and transform data

Our solution is composed of three pillars, each supporting specific data platform capabilities. The first pillar provides the capability to connect, ingest, and transform data.

There are four main ways to inject data into an architecture to enable manufacturing organizations to move from reactive to predictive maintenance.

• To start our process, we’ll enable the bulk transfer of operational transaction data. Bulk transfer services are used in situations where large volumes of data need to be moved to Oracle Cloud Infrastructure (OCI) for the first time—for example, data from existing on-premises analytic repositories or other cloud sources. The specific bulk transfer service used will depend on the location of the data and the transfer frequency. For example, OCI Data Transfer service or OCI Data Transfer Appliance may be used to load a large volume of on-premises data from historical planning or data warehouse repositories. When large volumes of data must be moved on an ongoing basis, we recommend using OCI FastConnect, which provides a high-bandwidth, dedicated private network connection between a customer’s data center and OCI.
• Frequent real-time or near real-time extracts are commonly required, and data is regularly ingested from warehouse management, scheduling, and order management systems using OCI GoldenGate. OCI GoldenGate uses change data capture to detect change events in the underlying structure of the systems that need to be serviced (for example, the addition of a new component, completed maintenance operations, changes in weather, and so on) and sends the data in real time to a persistence layer and/or the streaming layer.
• For manufacturing companies, analyzing data in real time from multiple sources can help provide valuable insights into their operational efficiency and overall performance. In this use case, we use streaming ingest to ingest all the data read from sensors through IoT, M2M communications, and other means. The ability to capture and analyze data streams in real time is critical to a manufacturer’s ability to perform predictive maintenance on their assets. Streams can originate from several ISA-95 Level 2 systems, such as SCADA systems, programmable logic controls, and batch automation systems. Data (events) will be ingested and some basic transformations/aggregations will occur before it is stored in OCI Object Storage. Additional streaming analytics can be used to identify correlating events and any identified patterns can be fed back (manually) for a data science examination of the raw data.
• To analyze this high-frequency streaming data in real time, we’ll use streaming processing to deliver advanced analytics. While traditional analytics tools extract information from data at rest, streaming analytics assesses the value of data in motion, i.e., in real time. And that’s not the only benefit. Because streaming analytics can be highly automated, it can help manufacturers reduce operating costs. For example, streaming analytics can provide real-time data on basic utility costs, such as electricity and water. Factories and plants can then use an automated streaming analytics tool to access instant insights regarding areas that could be optimized to reduce energy costs and respond appropriately to certain operational events using artificial intelligence. Streaming analytics can also make real-time predictions about upcoming equipment maintenance requirements, helping companies prepare well in advance for any upcoming repairs or routine upkeep.
• One potentially important optional component of the architecture is OCI Roving Edge Infrastructure, which can be used with Oracle Roving Edge Devices at remote facilities, such as a power plant or a solar panel field. OCI Roving Edge Infrastructure offers all the OCI services, so it can replicate a predictive maintenance architecture. OCI Roving Edge Infrastructure can also be used as a data hub for streaming data before it’s fed to streaming ingestion/processing in the cloud.
• While real-time needs are evolving, the most common extract from ERP, planning, warehouse management, and transportation management systems is some kind of batch ingestion using an ETL process. Batch ingestion is used to import data from systems that can’t support data streaming (for example, older SCADA or maintenance management systems). These extracts can be ingested frequently, as often as every 10 or 15 minutes, but they are still batch in nature as groups of transactions are extracted and processed rather than individual transactions. OCI offers different services to handle batch ingestion, such as the native OCI Data Integration service and Oracle Data Integrator running on an OCI Compute instance. The choice of service would primarily be based on customer preference rather than technical requirements.

Persist, process, and curate data

Data persistence and processing is built on three (optionally four) components. Some customers will use all of them, others a subset. Depending on the volumes and data types, data could be loaded into object storage or loaded directly into a structured relational database for persistent storage. When we anticipate applying data science capabilities, then data retrieved from data sources in its raw form (as an unprocessed native file or extract) is more typically captured and loaded from transactional systems into cloud storage.

• Cloud storage is the most common data persistence layer for our data platform. It can be used for both structured and unstructured data. OCI Object Storage, OCI Data Flow, and Oracle Autonomous Data Warehouse are the basic building blocks. Data retrieved from data sources in its raw format is captured and loaded into OCI Object Storage. OCI Object Storage is the primary data persistence tier, and Spark in OCI Data Flow is the primary batch processing engine. Batch processing involves several activities, including basic noise treatment, missing data management, and filtering based on defined outbound datasets. Results are written back to various layers of object storage or to a persistent relational repository based on the processing needed and the data types used.
• All data types stored in both raw and processed formats are persisted in a transactional data store. An Oracle database, such as Oracle Autonomous Database, works efficiently for all use cases, but some customers may think some data would be better stored on a NoSQL database. However, this isn’t necessarily true—an optimized Oracle database on top of Exadata will be faster than a NoSQL database for writes—and by using a NoSQL database, you lose out on the benefits of having all your data in a single repository with a unified view. You can also take advantage of hybrid partitions to keep only a few partitions in Exadata storage and the rest of the data in object storage as most operational analytics and dashboards use the most recent data.
• We’ll now use a serving data store to persist our curated data in an optimized form for query performance. The serving data store provides a persistent relational tier used to serve high-quality curated data directly to end users via SQL-based tools. In this solution, Oracle Autonomous Data Warehouse is instantiated as the serving data store for the enterprise data warehouse and, if required, more-specialized domain-level data marts. It can also be the data source for data science projects or the repository required for Oracle Machine Learning. The serving data store may take one of several forms, including Oracle MySQL HeatWave, Oracle Database Exadata Cloud Service, or Oracle Exadata Cloud@Customer.

Analyze data, predict, and act

The ability to analyze, predict, and act is facilitated by three technology approaches.

• Advanced analytics capabilities are critical for maintenance optimization. In this use case, we rely on Oracle Analytics Cloud to deliver analytics and visualizations. This enables the organization to use descriptive analytics (describes current trends with histograms and charts), predictive analytics (predicts future events, identifies trends, and determines the probability of uncertain outcomes), and prescriptive analytics (proposes suitable actions, leading to optimal decision-making).
• In addition to advanced analytics, increasingly data science, machine learning, and artificial intelligence are used to look for anomalies, predict where breakdowns might occur, and optimize the sourcing process. OCI Data Science, OCI AI Services, or Oracle Machine Learning can be used in the databases. We use machine learning and data science methods to build and train our predictive maintenance models. These machine learning models can then be deployed for scoring via APIs or embedded as part of the GoldenGate stream analytics pipeline. In some cases, these models can even be deployed in the database using the Oracle Machine Learning Services REST API (to do this, the model needs to be in Open Neural Network Exchange format). Additionally, OCI Data Science for Jupyter/Python-centric notebooks or Oracle Machine Learning for the Zeppelin notebook and machine learning algorithms can be deployed within the serving or transactional data store. Similarly, Oracle Machine Learning and OCI Data Science, either alone or in combination, can develop recommendation/decision models. These models can be deployed as a service, and we can deploy them behind OCI API Gateway to be delivered as “data products” and services. Finally, once built, the machine learning models can be deployed into applications that are part of a distributed control system (if permitted) or deployed at the edge via an Oracle Roving Edge Device or similar.

The multiple models created by combining data science with the patterns identified by machine learning can be applied to response and decisioning systems delivered by AI services.

• OCI Anomaly Detection can help monitor supply chain performance metrics (for example, raw material inventory, production throughput, work in progress, transit times, inventory turnover, and so on) in real time to identify and address disruptions. In a complex supply chain, the severity score of identified anomalies can help prioritize observed business disruptions for action.
• OCI Forecasting can help forecast supply chain metrics, such as demand, supply, and resource capacity, so appropriate actions can be taken to prepare ahead of time.
• OCI Vision and OCI Language can help understand documents, such as outgoing product quality reports and product defect reports, to enrich supply chain data.

The final yet critical component is data governance. This will be delivered by OCI Data Catalog, a free service providing data governance and metadata management (for both technical and business metadata) for all the data sources in the data platform ecosystem. OCI Data Catalog is also a critical component for queries from Oracle Autonomous Data Warehouse to OCI Object Storage as it provides a way to quickly locate data regardless of its storage method. This allows end users, developers, and data scientists to use a common access language (SQL) across all the persisted data stores in the architecture.

Use your data to improve manufacturing operations and increase profitability

With predictive maintenance, equipment is serviced only when it needs to be serviced, reducing unexpected outages. This delivers multiple advantages that include fewer scheduled maintenance repairs or replacements, the use of fewer maintenance resources (including spare parts and supplies), and, simultaneously, fewer failures. These proactive predictions can help to prolong the life of equipment while reducing the risk of potential product delays by minimizing equipment changeovers and the associated downtime.

Reducing unplanned downtime helps optimize business operations, improving efficiency, productivity, and speed and helping ensure the right part gets to the right place at the right time. Meanwhile, reducing maintenance, labor, and material costs and optimizing asset lifecycle costs increases profitability.