Seite wählen

Leider ist der Eintrag nur auf Amerikanisches Englisch verfügbar. Der Inhalt wird unten in einer verfügbaren Sprache angezeigt. Klicken Sie auf den Link, um die aktuelle Sprache zu ändern.

Work-related accidents and fatalities are a cause for global concern. According to the International Labour Organization, 612 workers have a work-related accident every minute. Over 1 million work-related fatalities occur each year. This article will explore the potential impact of Machine Learning for Predictive Maintenance and Industrial Analytics on workplace accidents and fatalities.

One of the basic tenets of O&M planning is the inverse correlation between Scheduled Preventive Maintenance and workplace accidents. There is a logic to this assertion: performing routine maintenance work identifies asset degradation before further breakdown occurs. At a macro-level, it is counter-intuitive to suggest that Preventive Maintenance may be the cause of unintentional accidents. The reality is more complex. In fact, Preventive Maintenance can lead to additional accidents and in some cases, fatalities.

There are several scenarios where Asset Maintenance is the catalyst (if not the cause) of safety incidents. These scenarios are based on the factors listed below:

Asset Age as a Risk Factor:  Aging infrastructure increases the need for both scheduled and unscheduled maintenance. In the US, the average age of industrial equipment has climbed to the level reached during World War II. In particular, the oil and gas industry is plagued by assets that are beyond their original design life. A report by Accenture suggests that over 50% of offshore platforms are older than their intended design life.

Industry 4.0 and the Average Age of Industrial Assets

Source: U.S. Bureau of Economic Analysis

The chart below represents the Bathtub Curve that maps the failure rate of an asset versus time. As assets wear out, additional maintenance work is necessary.industrial analyticsIndustry-Specific Influencers on the Incidence Rate: The level of risk varies by industry. The fatality rate for workers in mining, quarrying and oil and gas extraction is 11.4 per of 100,000 employees per year which is significantly more than the manufacturing industry fatality rate of 2.3 per 100,000. US Occupational Safety and Health Administration (OSHA) data for Severe Injury Rates for industries of more than 100,000 workers provide a more detailed view. Severe Injury RateSource: OSHA

Contributing factors to the delta between industries such as oil and gas and traditional manufacturing are the remote nature of work and the reliance on contractors that lack safety expertise. From a maintenance perspective, it is logistically more difficult to access isolated upstream locations than it is to service an industrial manufacturing plant.

The Inherent Risk of Human Error: By definition, there is an inherent level of danger to Asset Maintenance activities. Machines can be accidentally switched on, workers can be hit by a moving part or caught-in or compressed by equipment or objects.

Workplace accidents are the inevitable consequence of maintenance, no matter what age the asset. Of course, the incidence varies based on numerous factors, but maintenance activities carry an underlying risk.

The Role of Reactive Maintenance in Facility Safety

According to one study, an accident is 28% more likely to occur if maintenance work is reactive versus scheduled. Machine shutdown creates a more dangerous working conditions for employees. When a machine fails, the immediate objective is to avoid a domino effect that can result in an expensive facility shutdown and loss of production. This pressure adds additional stress to the maintenance teams. Furthermore, when the cause of failure is not known, the likelihood of mistakes occurring using trial and error increase. Finally, if machinery is not maintained adequately or in sufficient intervals, Reactive Maintenance becomes more dangerous.

The chart below is based on survey data and indicates a positive relationship between the Reactive Maintenance and safety incidents. Most respondents believe that more than 50% of incidents occur when maintenance is reactive.

Safety Incidents versus reactive maintenanceSource: IDCON Safety/Reactive Maintenance Survey 

The Role of Scheduled Preventive Maintenance

Intuitively, preventive maintenance reduces the likelihood of workplace injury. Research studies point to an inverse correlation between the Lost Workday Case Incidents (LWDCI) and maintenance. The logic is that if machinery is adequately serviced, they are less likely to breakdown and few people will be injured in the process.

At the same time, Scheduled Maintenance is not without risk. As with Unscheduled Maintenance, accidents occur during Scheduled Maintenance for reasons related to human error and poor safety standards. Below is a list of contributing factors to incidence occurrence:

  • Organization issues including chain of command
  • Insufficient employee training
  • Use of untrained subcontractors and temporary employees
  • Not conducting a maintenance risk assessment
  • Poor safety standards and procedures such as inadequate protocols to check completed work
  • Failing to identify hazards
  • Not having safe access to work area and escape route
  • Faulty maintenance equipment

There are numerous examples of avoidable fires and explosions that resulted from poor maintenance standards. For instance, the Stockline Plastics factory in Glasgow, Scotland exploded on May 11th, 2004. Nine workers were killed and 40 others were injured. According to academic research, the largest contributing factor to the tragedy was inadequate plant maintenance. For example, at one point, critical maintenance work was carried out by an inexperienced student.

It’s not only a lack of maintenance that leads to accidents. Sometimes, it is the wrong type of maintenance or too much redundant maintenance. For instance, at the Sodegaura Refinery in Japan, an explosion in October 1992 resulted in the death of 10 people. 7 more were injured. According to a research report, repeated ratcheting led to the reduction in the diameter of the gasket retainer that kept the heat exchanger airtight. Other maintenance errors included incorrect replacement of the gasket retainer and the removal of insulation that led to thermal deformation of the inner part of the tube.

Other instances of maintenance errors include the Bhopal Gas Tragedy of 1984 and the Phillips 66 Disaster of 1989.

The Era of Industrial Analytics for Predictive Asset Maintenance

Industry 4.0 formalizes concepts and best practices that already exist. Big Data, Machine Learning, and Artificial Intelligence are topics well-rooted in academia. The operational and business application of these disciplines is the underlying IP engine of Industry 4.0.

The application of Big Learning to Predictive Asset Maintenance is accelerated by the significant reductions in the cost to collect, store, transport and analyze sensor data. Since 2014, the cost per sensor has fallen by 60%. More dramatically, the cost to process data was reduced by 60X in the last decade.

Cost of Sensors

 Source: Goldman Sachs

 How Machine Learning for Predictive Asset Maintenance Can Lower Accident Rates

Industrial machinery is becoming more complex and maintenance activities can be the cause or catalyst for workplace injury.

How does Machine Learning change the Maintenance paradigm? At a high level, Industrial Analytics unlocks a key asset that is often overlooked: machine and sensor data and the intrinsic insights includes. In a McKinsey report, an example was given of an oil rig whose technicians only examined 1% of data generated by with 30,000 sensors.

Many of these sensors are in remote or dangerous places that are not easily or safely accessed. With advanced Machine Learning, the sensor data can be operationalized and asset degradation can be identified in advance of failure. Algorithms detect abnormal sensor behavior and alert facility owners in advance. With Machine Learning, fewer crews are dispatched to perform superfluous maintenance activities. By reducing the risk of human error, we lower the probability for maintenance-related accidents and fatalities.

Although still in its nascency, there are several options for Machine Learning for Asset Maintenance:

Manual Statistical Modelling: Sample data is taken from production assets and data scientists build statistical models. The data scientists extrapolate from the sample data to the specific machines. Manual statistical modelling has a low barrier of entry, but may not be scalable. Even with the most sophisticated algorithms, this approach is not based on a real-time analysis of machine behavior.

The Digital Twin: A virtual clone of a machine is customized based on the blueprints of the physical asset. The methodology used is referred to as Supervised Machine Learning. In real time, it monitors the physical asset and can provide alerts of upcoming failure. It is important that there be no discrepancy between the actual asset and the Digital Twin so that it can accurately simulate the operation. The Digital Twin is an expensive solution so it may not be economically viable to develop for machines that are towards the end of their original design life.

Unsupervised Machine Learning for IIoT Predictive Maintenance: This form of Machine Learning extracts and analyzes all sensor data from a production facility and provides alerts of machine degradation and upcoming failure. This methodology differs from the Digital Twin, insofar as the learning algorithm uses advanced artificial intelligence to build the machine models, detect abnormal sensor data patterns (or correlations of patterns) and then uses this information to provide real-time machine asset maintenance. Unsupervised Machine Learning is significantly cheaper than the Digital Twin approach because it does not require a simulated model to be built based on the machine blueprints and multiple engineers’ involvement.

Unsupervised machine learning for industrial analytics

Other Point Solutions: Vibration or acoustic monitoring can detect abnormal machine behavior that is indicative of pending failure. These solutions are not comprehensive but can be used for high priority or remote machines.

Conclusion: Creating the Optimal Mix of Maintenance and Analytics

Artificial Intelligence is not a solution in a vacuum. At its core, reducing workplace injury is about instituting best practices in Occupational Safety and creating a culture that prioritizes prevention.

We are not suggesting that Machine Learning will ever replace maintenance activities. Although asset maintenance exposes workers to a level of risk, failing to perform any maintenance activities is much riskier.  The key objective is Asset Maintenance Optimization. Incorporating Machine Learning for Predictive Maintenance into the overall Asset Maintenance mix can reduce the level of avoidable and redundant maintenance activities. The result is a safer workplace.

Are you considering Machine Learning for Predictive Maintenance?  If you wish to learn more about Presenso’s Machine Learning solution, please click here to schedule a complimentary demo.