top of page
Untitled design(155)_edited_edited.jpg

Maintenance of a Drone Fleet


ABSTRACT: Maintenance and logistics optimization was applied to a fleet of drones, operating from three sites with a central stock. The optimization achieved a Life Cycle Cost reduction of 34% while the fleet availability increased. This paper presents the optimization process, methods and results. Similar methods can be applied to a variety of other fleets.

1. Introduction

1.1 Motivation

The drone industry is one of the fastest-growing markets today (Forny & van der Meulen 2017). Drone failures pose both safety and financial risks, yet the drone failure rate is much higher than the failure rate of manned aircraft (Bone & Bolkcom 2003) Therefore, a great need exists for logistics and maintenance optimization of drone fleets.

In this paper, we present an example of modeling and optimizing the maintenance policy of a fleet of drones.

      1.2 Case Description

A fleet of 11 surveillance drones, operating from 3 different sites is considered (4 drones in site 1, 4 drones in site 2, and 3 drones in site 3). A central stock services the three sites.

Site surveillance is considered as not operational when more than 1 drone is failed. During such downtime, a penalty is paid by the drone operator to the site owner.

To optimize the fleet logistics and maintenance policy, the fleet operation had to be modeled. Following is a list of the parameters which were used to create a detailed model of the fleet behavior:

Reliability Data

  • Component failure distribution

  • Component failure modes

  • Drone redundancies

  • Operation profile

Maintenance Data

  • Component repair / discard policy

  • Repair time

  • Corrective maintenance

  • Preventive maintenance

  • Inspections

Logistic Data

  • Spare parts

  • Transportation times

  • Procurement time

Financial Data

  • Cost of spare parts

  • Penalties due to operation agreement

  • Corrective maintenance

  • Preventive maintenance

  • Inspections

Figure 1 presents the fleet breakdown tree. The fleet tree includes three main branches (one for each operation site), and under each branch, the drones and their components are described. This study focused on several drone sub-systems: Navigation, GPS, Inertial Measurement Unit (IMU), and the flaps.

The “Reliability Model” column in Fig. 1 describes the relevant model for each sub-system. For example:  the GPS sub-system includes two redundant GPS units (parallel model).

The “Distribution Type” column in Fig. 1 presents the failure distribution type for each component. Electronic components were assigned an Exponential failure distribution whereas the mechanical gyros and flaps were given a Normal distribution that describes their aging behavior.

Figure 1. Fleet breakdown tree.png
Figure 1. Fleet breakdown tree

2. Calculation Details


A commercial software (apmOptimizer) was used for the optimization. The software employs a combination of analytic methods (Birolini 1999) for calculating the fleet Life-Cycle Cost (LCC), and identifying cost and failure drivers. The analytic methods include:

  •  Markov chains for modeling spare parts supply, demand, and spare waiting times.

  •  Block mean failure rate calculations that account for component failure distributions, reliability models, scheduled maintenance, inspections, and the mission profile.

While analytic calculation is not as flexible as Monte-Carlo simulations, the analytic method is much faster. The speed of evaluating each model allowed for fast optimization of the maintenance and logistic policies using modified Dynamic Programming.

Dynamic Programming algorithms (Cormen, Leiserson, Rivest & Stein 2009) are ideal for bottom-up optimization of trees where the tree branches are independent.

However, in the fleet case, the branches are not completely independent: A central stock services the three sites, therefore a failure in one site affects spare part availability in the other sites. A modified dynamic programming algorithm was used to account for the inter-site dependencies. For example: The Markov chain model that describes the GPS voter spare parts supply and demand accounts for all 11 operating units, serviced by a single central stock.

The optimization goal is to achieve high reliability and availability while minimizing the LCC. Optimization was achieved by using the following optimization modules:


  • Optimal LOR: Level Of Repair Analysis – Optimization i.e. Repair/Discard policy. Repair is usually cheaper than buying a new component, however, long repair time (compared to procurement time) may require large and expensive safety stock. Therefore, the discard strategy is sometimes advantageous even when a repair is cheaper than buying a new component.

  • Optimal PM: Preventive Maintenance Optimization. Periodic maintenance is required for components that exhibit an aging behavior (failure rate increases with time) and cannot be inspected for degradation. In our case scheduled maintenance is relevant to the mechanical gyros.

  • Optimal PdM: Predictive Maintenance Optimization – inspections schedule. Periodic inspections are used to identify flap degradation. Beyond a degradation threshold, preventive maintenance is used to rejuvenate the flaps.

  • Optimal I: Inventory Optimization. Optimal I find the optimal combination of spare parts that minimizes the fleet LCC. The optimal spare part combination is as cheap as possible while ensuring a low probability of downtime due to stock out.


To emulate the case of under-maintained fleets, an initial maintenance policy was defined with few spare parts, inspections, and scheduled maintenance events. In each optimization step some maintenance actions / spare parts were changed to find the optimal combination.


3. Results


Fleet availability at each site as well as the fleet LCC were calculated at each step of the optimization process. Table 1 presents a summary of fleet availabilities at the various sites and the total LCC at each optimization step:

Screenshot 2024-04-16 132310.png
Figure 1. Fleet breakdown tree

It can be seen from Table 1 that the optimizations resulted in increased fleet Availability and LCC reduction of 34%. Optimal-LOR, Optimal-PM, and Optimal-I decreased the LCC but had a small effect on site availabilities. Optimal-PdM had a strong effect on both availability and LCC. This is not surprising since increased availability means a lower downtime financial penalty.

4. Conclusions


Fleet operators can reduce operation costs without jeopardizing performance by using analytic tools such as the apmOptimizer.

The modeling and optimization methods that were used in the example apply to any fleet, and are therefore relevant to many industries: defense, rolling stock, aviation, and mining.

Furthermore, the method is also good for modeling MRI medical machines, industrial printers, and other sets of identical machines that are operated at different sites and are maintained by the OEM.




Birolini, A. 1999. Reliability Engineering Theory and Practice, 3rd edition, Springer.

Bone, E. & Bolkcom, C. 2003. Unmanned Aerial Vehicles: Background and Issues for Congress. Report to Congress, Congressional Research Service, Library of Congress, pg. 2.

Cormen, T. H., Leiserson, C. E., Rivest, R. L., & Stein, C. 2009. Introduction to Algorithms, Third Edition. The MIT Press, Cambridge, Massachusetts, London, England. Pg. 359.

Forny, A. A. & van der Meulen 2017. R. Gartner.

 Learn More About Our Patented AI Reliability Solutions

BQR: Beyond Quality & Reliability
bottom of page