Choosing the right maintenance strategy is crucial for the efficiency, safety and longevity of non-intrusive inspection (NII) systems.

Today, equipment manufacturers and their service providers usually offer their customers the following maintenance models:

• Preventive Maintenance (PM), which relies on predetermined intervals that can be either before or after the equipment breaks down.
• Corrective Maintenance (CM), which fixes failures when they occur and is mainly a reactive “run until it breaks” strategy.
• Inventory, which requires continuously maintaining large stockpiles of spare parts for the purpose of performing unplanned corrective maintenance.

These reactive and preventive models present a number of problems. By conducting unnecessary maintenance tasks and emergency repairs, customers usually spend 100% of the equipment’s purchase costs on maintenance during the first 10 years. Moreover, equipment downtime can be high. The consultancy firm Roland Berger estimates that approximately 30% of preventive maintenance efforts are unnecessary. According to another consultancy firm, Emerson Process Management,[1] in 30% of cases, they are counter-productive, with as many as 70% of failures occurring shortly after major maintenance actions. Furthermore, over 90% of the failures typically result from conditions that can occur at any time and quite unexpectedly.

Transforming maintenance through AI and IIoT

Today, another maintenance strategy called condition-based maintenance (CBM) is used to replace preventive maintenance for critical equipment components. The maintenance activities are planned based on the technical condition rather than a predefined interval. The health technical condition is determined by a service technician, who remotely assesses the health of the equipment using sensor data and real-time monitoring tools. This strategy can be enhanced by applying machine learning (ML) and artificial intelligence (AI) algorithms, known as predictive maintenance (PdM). It provides additional valuable insights that can be used not only to improve the scheduling of maintenance activities, but also to estimate the remaining useful life of critical equipment.  This has enabled many industries to reduce their maintenance costs by up to 40%,[2] while increasing equipment availability, extending asset life, reducing waste production and improving safety protocols.

Predictive maintenance combines domain knowledge, Machine Learning (ML) and AI algorithms to process technical data in near-real time and provide valuable insights into the technical health of the most critical components.

Applying predictive maintenance to NII systems requires using data provided either by system logs and existing built-in sensors, or by installing additional Industrial Internet of Things (IIoT) sensors. These IIoT sensors continuously collect real-time data for key technical parameters of the most critical components such as X-ray generators, tubes and accelerators. These data are then analysed using different algorithms to identify patterns, detect anomalies and predict failures.

Based on the equipment and customer needs, different types of predictive models can be developed to process and analyse data relevant to the equipment’s technical performance, including:

• Short-term predictive models – these focus on responding to near-future operational changes, typically within a time frame of a few days to a few weeks. These models are particularly valuable for identifying and responding to unexpected anomalies in real time. For example, they can detect abnormal behaviour of the air-conditioning system in a high-energy X-ray system, either in the detector line or in the technical (server) room, long before a critical pre-defined (CBM) threshold is reached.
• Mid-term predictive models – these extend the predictive analysis to address issues that are likely to develop over a period of a few weeks to a few months. These models relate to parts that are likely to fail due to wear and tear or ageing. For example, in low-energy X-ray systems, these models can predict a generator failure up to three months in advance by monitoring its electrical parameters.
• Long-term predictive models, also known as survival models – these focus on strategic planning and ensuring the long-term reliability of the equipment. These models aim to optimise the availability of spare parts and resources while minimizing the costs. The time frame for these models typically ranges from a few months to a year. For example, survival models can predict the failure rates of magnetrons or thyratrons over a long period of time, allowing optimization of spare parts to avoid both overstocking and stockouts.

While predictive maintenance uses data analytics to anticipate equipment failures before they happen, prescriptive maintenance (PxM) goes a step further by offering actionable case-by-case recommendations to support service specialists in fixing system issues. PxM ensures fast reaction and decision-making mechanisms for resolution management and repair.

All the data, trends and actionable insights generated by the system need to be brought together and displayed on a business intelligence (BI) dashboard. This BI dashboard should be able to provide service specialists and end users with a clear understanding of the asset’s health and the steps required to maintain peak performance.

In addition to the BI dashboard, a skilled service team operating from a maintenance control centre should be set up, as well as a 24/7 service desk and hotline support to address potential failures quickly and efficiently.

Other services to be provided should include:

• real-time remote monitoring of the equipment’s technical condition;
• remote diagnostics and troubleshooting;
• on-site interventions and provision of the necessary spare parts;
• radiation safety measurements in line with the applicable law depending on the type and configuration of the particular X-ray system used.

The initial investment to implement such a solution is usually around 1,000 euros per NII system, covering the hardware (IIoT device and sensors) and installation costs. However, depending on the preferred subscription model, these upfront costs can be deferred or included in the ongoing payments, effectively requiring zero initial investment.

Possible variants of subscription models are:

• pay per use;
• customer-tailored product packages and bundles; and
• value-based pricing models and price metrics.

Such flexibility enables Customs administrations to adopt predictive maintenance solutions without facing significant financial barriers.

It is worth mentioning that implementing predictive maintenance will also reduce overhead costs, such as those of travelling to and from the operational site for diagnostics and problem resolution and those associated with the storing of spare parts.

To protect critical data from cyber threats and vulnerabilities, it is necessary to develop a defence-in-depth strategy. Only technical information about the health of the NII system, such as logs from the system and data from sensors, should be collected, and the information should be stored in a maintenance control data centre, while ensuring that encryption is used at every stage of data transmission. Additionally, next-generation firewalls must be used to process data flows and block potentially dangerous traffic. Further security layers include intrusion detection systems, as well as real-time monitoring and incident response mechanisms to quickly identify and mitigate potential threats. Regular security audits, penetration tests and vulnerability scans must be conducted to ensure that the solution remains resilient to emerging cyber threats.

NII system vendors and Customs administrations should test predictive maintenance solutions in real-world conditions. They can also request proof-of-concept (PoC) projects to be conducted free of charge for the specific types and models of X-ray systems they use.

More information
www.danlex.bg / www.pm4x.eu
pm4x@danlex.bg

[1] White paper: Reducing Operations & Maintenance Costs, Emerson Process Management 2003.

[2] Operations & Maintenance Best Practices, Federal Energy Management Program, August 2010.