Your Predictive Maintenance Capabilities will be Enhanced by Big Data Technologies

Author: John Soldatos
Category: Enterprise Maintenance and Reliability

In most cases, maintenance activities are based on preventive measures, which focus on regular maintenance of machines, components and other infrastructure elements in order to prevent them from failing.

However, these preventive maintenance measures are not always optimal, as they usually perform maintenance earlier than needed, which reduces Overall Equipment Efficiency (OEE).

As a result, infrastructure operators and maintenance engineers are increasingly considering a shift towards predictive maintenance. This approach promises cost savings over routine or time-based preventive maintenance, because tasks are performed only when warranted. Implementation of predictive maintenance practices thus leads to optimized maintenance activities.

Predictive Maintenance (PdM): The Benefits

Many plant operators consider predictive maintenance the ultimate maintenance vision, providing many benefits, including:

Overall, predictive maintenance will minimize planned downtimes, while leading to better management of employees’ time and effort.

Barriers to Adoption of PdM

Despite the acknowledged benefits of predictive maintenance, its adoption is still in the early stages because the successful deployment is particularly challenging from both a technical and an operational perspective. Some of the main challenges include:

  1. The need to combine multiple maintenance detection modalities and associated datasets.
    Predictive maintenance hinges on processing multiple datasets, notably those associated with different sensors and maintenance detection modalities such as vibration analysis, oil analysis, thermal imaging, acoustics and more.

    While a single modality (e.g., vibration) can be used for predictions, the consideration of holistic datasets from multiple modalities leads to more stable and accurate predictions. However, the combination of multiple sensors requires additional investments in sensors and data collection and generally doesn’t happen, even in state-of-the-art maintenance systems.

  2. Data fragmentation.
    Even when data from multiple sensors are available, they tend to be isolated from each other in disaggregated “data islands.” This means that data resides in different systems (e.g., in CMMS like IBM Maximo or dedicated sensor databases) and are represented in diverse formats, which makes their integration challenging.

    Semantics across platforms and systems tend to be different, which is yet another factor that hinders their unified processing toward predicting when a machine or component will fail.

    Furthermore, there is no easy way to leverage additional data (e.g., quality of the measurements of a metering system, health data of an SCADA or Supervisory Control And Data Acquisition system), which could possibly enrich sensor data toward more accurate predictions.

  3. Lack of proper analytics algorithms and tools.
    The predictive analysis of large datasets requires advanced algorithms and tools beyond baseline machine learning and statistical models. Such algorithms are not widely available and their deployment and consequent business analysis require the engagement of skilled data scientists who are high in demand but low in supply.
A recent report from Mckinsey looked at data coming from approx. 30,000 sensors on an oil rig and found that 99% of the data was unused. The 1% (or less) of data used was mostly to detect and control anomalies and not for optimization and prediction, which provide the greatest value. (img source: Agência Brasil)

Big Data Technologies to the Rescue

The advent of Big Data technologies provides the means for overcoming the above challenges.

Big Data is generally an overhyped term which is sometimes used as a marketing pitch but predictive maintenance provides an ideal business case for deploying Big Data technologies. Big Data is about developing and deploying distributed, data-centric systems that extend the capabilities of state-of-the-art databases in order to handle datasets featuring the four Vs:

By handling the four Vs, Big Data technologies alleviate the data challenges of predictive maintenance through facilitating the unification, integration and real-time processing of very large, maintenance-related datasets.

They pave the way for predictive analytics that can provide credible insights on the condition of equipment and subsequently facilitate the anticipation of failures.

Predictive analytics are also propelled by the rise of Deep Learning technologies. These provide the means for processing not only numeric sensor data, but also multimedia data, such as imaging data and information from acoustic sensors.

Deep Learning goes hand in hand with Big Data, as it is usually deployed in conjunction with BigData infrastructures and technologies.

BigData is vital to successful PdM adoption and implementation

BigData technologies are certainly key elements in realizing the transition from preventive to predictive maintenance based on an optimal exploitation of available datasets.

They can derive insights on the condition of the equipment, including hidden patterns of maintenance. However, they can go much further to realize the full potential of predictive maintenance. Future predictive maintenance systems will be able to close the loop back to the plant floor through actions such as configuring devices or even stopping engines.

Such actionable intelligence will be achieved based on the integration of Big Data technologies with industrial automation.

Overall, the BigData revolution will enable plant operators and maintenance experts to complete the transition from preventive to predictive maintenance.

This evolution will not be only a matter of technology deployment, but also a matter of investment in complementary assets, such as new maintenance processes and employee training. The era of BigData analytics and predictive maintenance is fast approaching and we should be taking steps to prepare for the transition.

Author: John Soldatos

John Soldatos holds a Phd in Electrical & Computer Engineering. He is co-founder of the open source platform OpenIoT and has had a leading role in over 15 Internet-of-Things & BigData projects in manufacturing, logistics, smart energy, smart cities and healthcare. He has published more than 150 articles in international journals, books and conference proceedings, while he has authored numerous technical articles and blogs posts in the areas of IoT, cloud computing and BigData. He has recently edited and co-authored the book “Building Blocks for IoT Analytics”.

Similar Posts