Drones in manufacturing: exploring opportunities for research and practice

2.1 Drones as an advanced manufacturing technology

An AMT is a technological system that consists of hardware, software and support processes (Bessant and Buckingham, 1989; Cagliano and Spina, 2000; Chung and Swink, 2009) (see Figure A1, for a high-level architecture of a drone system). The hardware of a drone system includes the aircraft, the remote controller, the installed payload (cameras, sensors, carriers, etc.), local navigation support system, energy supply system and the information technology (IT) infrastructure. The software consists of the programs and algorithms that control the drone’s flight and payload tasks, as well as communicate with the controllers, navigation system and IT systems. The support processes cover a range of drone operations that involve humans, such as manual piloting or establishing and maintaining the infrastructure required for automatic or autonomous flights, and interpreting the data collected by the drone.

An AMT application can be “stand alone” or it can be linked or integrated with other technologies (Meredith and Suresh, 1986; Small and Yasin, 1997; Small, 2007). Most current drone applications are stand-alone designed to do a specific task. Examples include the inspection of hard-to-reach equipment using video and thermal cameras in the oil and gas industry, aerial photogrammetry for developing three-dimensional modeling during factory planning and delivering spare parts during maintenance operations (see Barth and Michaeli, 2018; Maghazei and Netland, 2018; ZF Friedrichshafen AG, 2018). Linked AMTs conduct their own tasks, but also communicate and coordinate with other technologies. An example of a linked drone system is applications used in inventory cycle counting to automatically update inventory records in a warehouse management system. Integrated AMTs are dependent on the tasks of other technologies. Only very few current drone applications qualify as integrated AMTs.

2.2 Physical vs analytical capabilities of advanced manufacturing technologies

AMTs can be classified according to their physical and analytical capabilities (e.g. Bessant and Buckingham, 1989; Kotha and Swamidass, 2000; Kotha, 1991; Steenhuis and Pretorius, 2016). The physical capability of an AMT is its ability to conduct physical tasks. For example, RFID readers and sensors have relatively low physical capabilities and CNC machines, conveyor systems, AGVs and industrial robots have relatively high physical capabilities.

Following Kotha (1991) and Kotha and Swamidass (2000), we define the analytical capability of an AMT as its ability to process data. What separates high from low analytical capability is not necessarily associated with tedious programming but is decided by the degree of data processing during use. For example, a simple special-purpose CNC machine has a high programming setup, but the machine usually simply runs its program during use and is therefore an AMT with low analytical capability. An advanced FMS, however, needs high analytical capability to synchronize the real-time flows between machining processes, machine tools and materials.

The capabilities of AMTs are improving quickly due to the advancement of complementary technologies such as sensors, robotics, cloud computing, big data and analytics, to name a few (Guo and Qiu, 2018; Frank et al., 2019). Such technological advancements are improving both the physical and the analytical capabilities of AMTs. For instance, Steenhuis and Pretorius (2016) discuss that the growth of consumer 3D printing is due to the advances of both its physical capabilities (e.g. extruder movement and speed), as well as its analytical capabilities (e.g. using open source software packages). In a study of the evolution of automated guided vehicles (AGV) in FMS, Buyurgan et al. (2007) show how the physical capabilities of AGVs are enhanced with analytical capabilities such as autonomous travel without guided paths.

The separation of analytical capabilities vs physical capabilities of AMTs makes intuitively sense for drone systems. A drone with low analytical capability only captures input data without processing it, examples are simple photography or filming. A drone with high analytical capability converts the input data to other forms of data or information. For instance, a drone system that is equipped with a thermal camera receives input data and produces thermal images of temperature radiation. A drone with low physical capability is not able to perform any physical operation other than flying. A drone with high physical capability has the ability to conduct physical tasks such as move objects (in addition to the mounted payload) (e.g. parcel-, part- or tool delivery), or to perform a physical operation in addition to flying (e.g. spraying chemicals or repairing scratches).

Using the separation of analytical capabilities vs physical capabilities, we derive a conceptual framework from the AMT literature (Figure 1). The combination of low and high for these two capabilities suggests that AMTs can be classified into four different types, respectively, low-low, high-low, low-high, and high-high configurations of analytical vs physical capabilities.

A conceptual framework explains graphically “the main things to be studied” and can “evolve as the study progresses” (Miles et al., 2014, p. 20). We use the theory-informed framework in Figure 1 as a conceptual underpinning when we go to the field to study the applications of drones as a new form of AMT in manufacturing.

3.1 Selection of experts

We searched for informants who possessed technical knowledge, procedural knowledge and interpretive knowledge in the field of industrial drone applications (Bogner and Menz, 2009). These experts were required to have first-hand experience in applying drones in industrial settings because they would then be “in a position to actually put their own interpretations into practice” (Bogner et al., 2009, p. 7). Identifying experts using objective measures (such as years of experience or number of flight hours) was infeasible because of the unavailability of data. Therefore, we selected informants using the purposeful sampling strategy combined with intensity sampling, reputational sampling and snowball sampling (Patton, 2015). First, we visited locations where drone experts gather, such as drone exhibitions, technical conferences and drone seminars. Second, we searched the internet for media articles, white papers and commentaries written by, or mentioning, drone experts. Finally, we used word-of-mouth to access the network of drone experts whom we had already interviewed.

We focused the data collection procedure on drone vendors and drone service providers. Although we searched intensively for informants about cases of drone implementation in manufacturing, we could only identify a few companies that had run experiments with drones in that industry. Moreover, most of these companies were reluctant to share their experiences, which we speculated was for reasons of confidentiality, concerns about intellectual property rights or the failure of experiments.

3.2 Data collection

The expert interviews were conducted during the European Drone Expo held in March 2017 and March 2018 in Brussels, Belgium; the UAV Expo in June 2017 in Brussels, Belgium; and the UAV Expo in April 2018 in Amsterdam, the Netherlands. Other interviews were carried out in Switzerland and Germany, and a few were conducted in video conferences (see Table I). Almost all the expert interviews were conducted face-to-face, which was recommended by Christmann (2009). In total, we conducted 66 interviews with drone experts from 56 companies headquartered in 13 countries worldwide. We decided to stop conducting new interviews when we reached the saturation of information (Eisenhardt, 1989). The companies included drone manufacturers, drone service providers, drone insurance companies, regulatory bodies, training centers, software developers and part suppliers. Typical in this industry, most of the companies were small- and medium-sized, which enabled us to communicate with C-level executives and the founders of many of the firms.

We conducted semi-structured interviews. We first designed an interview guide based on the guidelines of Bryman and Bell (2015). It was organized into four different topics: evaluation of drone applications, implementation of drone applications, results of drone applications and lessons learned from experiments and implementation. We used the same version in the interviews conducted between March 2017 and March 2018. After conducting more than half of the interviews, we conducted a preliminary analysis and reduced the survey to questions that had not yet reached saturation in their responses. We used this second version in the interviews conducted after March 2018. Versions of both the early and the late interview guide are included in Appendix 2.

The experts’ perceptions of the interviewer’s technical, procedural and interpretive knowledge could affect the quality of the interviews (Bogner and Menz, 2009). Because we are experts in manufacturing, we qualified as an “interviewer as an expert from a different knowledge culture.” However, we were laypersons regarding drones. During the initial interviews, we therefore assumed the role of a “knowledgeable layperson” (Froschauer and Lueger, 2009). We researched drone technologies, attended introductory courses for drone pilots, attended drone exhibitions and seminars and acquired a set of drones. As the interviews progressed, we became accustomed to drone technology, terminology and applications. After conducting about 25 interviews, we classified ourselves as “co-experts” (Bogner and Menz, 2009).

To increase the validity of our findings, we collected complementary data from other sources. For example, we analyzed 39 keynote presentations and 28 recorded keynote talks given at the first and second FAI International Drones Conference and Expo in Lausanne in Switzerland in 2017 and 2018, respectively. We also tape recorded 13 keynote addresses and panel discussions at the UAV Expo 2018 in Amsterdam. Another important source of data was obtained in our review of 474 pages of catalogs and manuals from over 35 different companies, which we collected from the exhibitions, seminars and visits. We also reviewed 57 white papers and industrial reports amounting to 830 pages of text; we read more than 100 media articles; and we viewed online videos about drone technology and its applications.

3.3 Data analysis

We followed the six-step analytical approach to analyzing expert interviews, which was recommended by Meuser and Nagel (2009): (1) transcription, (2) paraphrasing, (3) coding, (4) thematic comparison, (5) sociological conceptualization and (6) theoretical generalization. This approach is similar to thematic analyses aimed at identifying, analyzing and reporting patterns (themes) within data with high flexibility, which allows researchers to interpret various aspects of the research topic (Braun and Clarke, 2006). Steps 1 through 4 involve objective processes, whereas Steps 5 and 6 rely on subjective interpretation and ingenuity. As usual in explorative qualitative research, this process was iterative (Eisenhardt, 1989).

First, we transcribed 47 of our interviews and took notes on 19 interviews, which resulted in more than 600 double-spaced pages of raw text. We were not able to tape record and transcribe 19 interviews mainly because of the noisy environment and in a very few cases, because no tape recording was allowed. For these exceptions, we took notes during and immediately after the interviews. All interview recordings and notes were stored in a research database. Second and third, we used the software package ATLAS.ti to paraphrase and code the text from the interviews by using the closest possible language to the data (Strauss and Corbin, 1990). In paraphrasing, the text was sequenced according to thematic units. The open coding resulted in a list of 409 codes, of which 142 codes described use cases that corresponded to the first part of the interview guide. In the fourth step, we performed thematic comparisons by classifying similar codes into 57 empirical themes.

In the fifth step, we started to extract new knowledge from the sorted and coded data. In sociological conceptualization, “the specific characteristics of the commonly shared knowledge of experts are condensed and categorizations formulated” (Meuser and Nagel, 2009, p. 36). We aggregated the empirical themes into 17 conceptual categories. The final step was to “arrange the categories according to their internal relations” (Meuser and Nagel, 2009, p. 36) seeking generalization by developing a typology of industrial drone applications that builds on the conceptual framework developed through the review of the literature. A typology is a “conceptually derived interrelated sets of ideal types” (Doty and Glick, 1994) that assists the theoretical progress of a field by breaking it down into subparts with distinct characteristics (Miller, 1996). After developing the typology, we drew on the rest of the data from the interviews to discuss both opportunities and challenges in drone applications.

4.1 Industrial applications of drones

Table II shows our data reduction structure for the industrial applications of drones (e.g. Gioia et al., 2013; Ramus et al., 2017). The left column shows the empirical themes that describe the current applications of industrial drones, which directly emerged from the transcription, paraphrasing and coding of interview data. The middle column shows the conceptual categories that emerged from the thematic comparison of the empirical themes. We summarized empirical themes (i.e. drone applications) into generic concepts (i.e. areas of application) based on similarities through a process of abstraction (Flick, 2014, p. 404). For example, drone applications for gas detection and noise monitoring were summarized into the area of “hazard identification.” The right column shows our theory-informed generalizations of activities that a drone can provide: see, sense, move and transform. We derived the four aggregate dimensions by elaborating on the relations between concepts through further abstraction. For instance, we aggregated all the application areas of drones in relation to visual capabilities into “see.”

“See” is the capability of collecting visual data, often in the forms of images and videos. In the manufacturing industry, examples are the visual inspection of equipment, such as gas flare, silos, boilers, drums, tanks, chimneys and pipelines (both above and below ground). These are common tasks in many process industries (e.g. petrochemical industry, offshore and onshore oil platforms). Drones that “see” are also used to monitor the safety of staff, such as during maintenance operations where fixed cameras are not economically feasible. Some large plants apply drones to monitor security instead of closed-circuit television or human patrols. Drones are also tested in applications used to monitor of safety, ergonomics and regulatory compliance.

“Sense” is the capability of collecting data and transforming it into the other forms of data or structured data (i.e. information) without performing additional physical operations. Some relevant examples in manufacturing include the following: the thermal inspection of equipment, machines, chimneys and stacks; gas detection and noise monitoring to identify hazards in the oil, gas and petrochemical industries; non-destructive tests such as measuring the thickness and detecting corrosion of equipment; cycle counting, tracking and trace and finding lost pallets and slots for inventory management; 3D factory planning and process mapping for the optimization of factory layouts and material flows.

“Move” is the ability of a drone system to grasp and carry objects or perform physical operations (e.g. spraying). A typical example in manufacturing consists of intra-logistics operations, such as delivering light components, spare parts or tools especially during maintenance operations. Drones can also be used to spray paint on the corrosion in equipment and buildings and to spray foam during fires.

“Transform” is the ability of a drone system to collect data and transform them into information while performing physical operations (e.g. carrying objects). It combines the capabilities of see, sense and move. Current examples of “transform” in industry are scarce, but a few promising pilot studies are underway. For instance, a drone system with a camera can simultaneously inspect equipment and perform simple repair operations using mounted tools (e.g. patching, painting and sealing). Drones can perform pick up operations in a warehouse. Both examples are technically complex and not economically feasible in the current state of the technology. For example, in e-commerce warehouse management, order picking and order sorting require advanced drones that grasp items and carry them reliably. This operation also requires multiple sensors (e.g. barcode-, data matrix- or RFID readers) to manage inventory and update warehouse management systems in real time. An efficient operation would require a swarm of autonomous drones with the capability of recognizing obstacles and applying avoidance algorithms.

Building on the theory-informed classification of AMTs in Figure 1, we can now use the empirical findings to propose a typology of drone applications in manufacturing. It is illustrated in Figure 2. Seeing is a low analytical and low physical capability. Sensing involves a high analytical capability and low physical capability. Moving represents high physical capability and low analytical capability. Transforming requires high analytical and high physical capabilities. We use this typology to discuss the current state of drone applications in manufacturing, propose a research agenda, and propose implications for practitioners.

4.2 Potential benefits of drones

We asked all interviewees about the potential benefits of using drones in manufacturing. Although the real benefits are related to specific-use cases and contexts, the data analysis showed that the potential benefits fell into five broad categories:

1. cost savings;

2. task speed;

3. safety improvements;

4. efficient data collection; and

5. public relations (PR) and marketing.

First, drones can increase productivity and hence reduce the costs of manufacturing. In particular, in manufacturing plants in inspection-intensive process industries, drones can bring a significant cost saving. Inspections carried out by drones reduce the amount of labor-intensive work and eliminate the need for scaffolding. Regarding an extreme non-manufacturing example, one interviewee reported that in an inspection project on one of the biggest oil platforms in the North Sea, the introduction of drones reduced a 700 person-days inspection of 14 objects to 28 person-days. Furthermore, the inspection of flare exhausts required a shut down in which time was an extremely precious resource valued at $7m per day. Another frequently reported example was the use of drones to count stocks in large warehouses. The cost savings in this application were derived from replacing human work, eliminating rework due to human errors and improving order fill rates, thus increasing customer satisfaction and decreasing safety stock levels. Similar findings were reported by Hoffmann (2017), in which an estimated annual operating cost savings of $300,000 was derived in scanning 1,000,000 barcodes per year in a warehouse of 500,000 square feet.

A related potential benefit is the increased speed of performing tasks. Using drones for the inspection of hard-to-reach equipment and installations speed up the operations because of the shorter setup time and higher maneuverability compared to traditional processes involving scaffolding, ladders and rope access. Shorter setup times and higher maneuverability can also increase the frequency of inspections, allowing for the faster detection of incidents, such as gas leakages. Another example is the use of drones for the inventory management of bulk raw material, in which light detection and ranging (LiDAR) scanning with drones can increase the speed and efficiency of inventory counting compared with handheld scanners. Another example was provided by an interviewee who explained that drones can speed intra-logistics operations:

Imagine an assembly line in the automotive sector where parts are not working or are missing. The normal process is then that a human being is running or biking to get the part from the warehouse. This could take between 10 and 15 minutes in normal cases. With the drone, you could fly over infrastructure and by that, you could do it in 3 to 4 minutes. These are real numbers we measured in an automotive factory.

Safety improvement was the most frequently mentioned benefit of drones. According to the president of a global drone association and the CEO of a drone start-up, “Dull, dirty, and dangerous, those are the jobs that drones improve on.” Drones can reduce hazardous tasks in many operations. In particular, drones can replace manual human inspection of hard-to-reach equipment and hazardous areas. Moreover, drones can be a supportive tool in conducting health, safety and environment (HSE) activities, such as sniffing for contamination and gas leaks or search and rescue operations during emergencies in large manufacturing plants. Drones can also film emergency drills to improve the responsiveness of HSE teams during evacuations.

A fourth benefit is that drones can increase data collection efficiency and assist acquisition of data that has not been collected before. For example, one interviewee explained, “A drone can get high quality, more consistent, and repeatable datasets, and that’s important because if you inspect the same structure many times, you see trends.” This capability is particularly promising in maintenance operations in process industries. Drone users can also increase the capability of data collection using multiple sensors. For example, drones can be used to provide digital 3D models of factory floors to support layout planning and redesign (Barth and Michaeli, 2018; Melcher et al., 2018). In general, the increased amount of accurate data collected by drones can be used to support managerial decision-making. An interviewee shared, “We use a drone to inspect and with all that data you can make decisions on what you do. Do I fix, do I inspect it again, or do I do nothing?” Drones that include complementary software packages for data analysis can provide decision makers with meaningful reports in easy-to-understand formats.

A fifth and more subtle benefit of drones is their use in PR stunts. Media outlets and newspapers have been quick to report on pilot studies of drones in factories. Consistent with the findings of previous works, press coverage and media attention is usual for companies that are early adopters of robots and other AMTs (Meredith, 1987). A few recent examples of press coverage for drones are reports of applications used in cycle counting in Mercedes warehouses (e.g. Banker, 2016), intra-logistics applications in ZF Friedrichshafen (Dellinger, 2018) and inspections of hard-to-access equipment in a Ford factory (Hatt, 2018), the Pilsner Urquell brewery (Margaritoff, 2018) and Royal Dutch Shell’s oil and gas facilities (Castellanos, 2018). Companies that use drones may be perceived as innovative and future-oriented, which can have positive effects on recruitment, public goodwill and brand value.

4.3 Challenges for drone applications

We identified five generic categories of challenges and drawbacks related to the use of drones in manufacturing:

1. technological challenges;

2. operational challenges;

3. organizational challenges;

4. legislative challenges; and

5. societal and mental challenges.

It is not surprising that major challenges to the industrial application of drones are related to technological limitations, the most frequently mentioned of which related to constraints in current battery technologies. The limited battery capacity implies that drone users must balance flight endurance with payload. As of 2019, commercially available industrial drones have a flight time between 2 and 25 min. After the mission, the batteries must be replaced or recharged. Recharging often takes 45 min or longer. One interviewee observed, “If the battery technology gets better or we can find a way to make something lighter, then we have more range.” Another solution is to eliminate the need for batteries by using tethered drones, which have a direct power supply and use wired data transmission. Tethered drones are a promising solution in applications that require high flight endurance and low hovering capabilities (e.g. inventory management). Other technological challenges include indoor navigation, reliable data transfer and communication, danger of explosion, safety mechanisms and noise. For example, indoor drones may need a combination of positioning systems, object recognition and collision avoidance algorithms, SLAM algorithms, as well as a combination of multiple sensors, including an inertial measurement unit.

The second challenge relates to the operation of the drone. Most current drone applications are manual pilot operations that are flown within the line of sight. The alternatives are automatic or autonomous flights. All operation modes pose a range of challenges. Manual operations require alert and skilled pilots. In long operations, pilot fatigue can quickly become a source of human error. The gates, doors, pillars, ventilation, fire protection installations, cranes, utility gateways and large machines in factory environments are challenging to navigate even by experienced pilots. Automatic and autonomous flights require a continuously maintained navigation infrastructure. In both cases, drone flights need to be reliable and safe, especially around people. Redundant systems, such as parachutes, extra propulsion and safety algorithms in autonomous flights, can make drones failsafe. Furthermore, the current drone technology is a poor fit in factory environments that are at risk for explosions or are sensitive to electrostatic discharge.

The organizational challenges include the need for skilled drone pilots, who not only must be able to fly drones safely but also must have a deep understanding of the tasks and missions involved. Human issues such as workers’ knowledge and technical experience, training and involvement in planning are key determinants for the success of technology adoption (Chung, 1996; Walton, 1987; McCutcheon and Wood, 1989; Pagell et al., 2000). The data collected in the interviews revealed that human error is a greater problem than technological error in drone operations. For example, pilots need to be trained in the use of a drone as an inspection device as well as to collect and deliver useful data. However, the use of autonomous drones may overcome the challenge of training drone pilots and keeping them alert. According to one interviewee, “autonomous drones are safer than drones with human pilots.” Other organizational challenges of adopting drones are related to developing a convincing business case that provides an acceptable return on investment. This is similar to the debate on measurable benefits of adopting AMTs in manufacturing industries (Swink and Nair, 2007; Udo and Ehie, 1996). On one hand, it is difficult to specify the potential savings that drones can provide in manufacturing. The costs, on the other hand, are visible to everyone. Therefore, risk averse managers often do not invest beyond trials. Yet, the results from trials can help managers to set expectations and to develop a risk profile for drone programs in their settings (Hottenstein and Dean, 1992). Furthermore, organizations that plan to invest in drone operations face the “make-or-buy” dilemma. The data collected in the interviews indicated that this decision should depend on the availability of internal and external skills and the sensitivity of both the processes and the data. An additional challenge concerns dealing with the data that are collected. In many firms, drones are only a small part of their “digital transformation.” Often, the use of drones must wait for the slow preparation involved in digitalization and data management. The co-founder of a leading drone service provider shared, “It’s not that drones can’t fly, it’s the fact that digitization of these big industries is difficult and lengthy.” This has also been pointed out in the AMT literature, which suggests a prevalence of stand-alone AMT applications and islands of automation with limited integration (Sun, 2000).

The fourth challenge concerns legislative rules and regulations. Although the number of drone applications is increasing, the regulations concerning their use is lagging. A main benefit of using drones in indoor applications is that the regulations are more relaxed compared with outdoor applications. There are large variations between countries in terms of drone legislation. The licenses (or the lack of them) define how, where and what applications the manufacturer can use drones. As in many emergent technologies, it has been difficult to regulate drones, which is because of the rapid improvement of the technology, safety and security issues, the lack of clarity of who should draft the regulations and the lack of knowledge about many real applications (Khanna, 2018). For instance, flying beyond visual line of sight (BVLOS) is prohibited in many countries, which reduces the applications of drones as well as the areas of coverage in outdoor applications; however, some countries make exceptions for flying BVLOS.

Finally, there are societal and mental challenges related to the use of drone applications in manufacturing. For example, the common use of drones as a military weapon affects public opinion. Many members of the public have negative perceptions of drones as a new technology. People are also concerned about the safety of drone technologies, the intimidating appearance and noisiness of drones and the invasion of personal data. In a case in Australia, drones were used to monitor staff behavior, but the practiced was stopped because it violated workers’ privacy (Opray, 2016). In operations that use heavy drones or payloads, safety concerns are justified.

Interestingly, only two of our 66 interviewees mentioned price as a drawback. This finding was surprising. Only few years ago, price would have been a major challenge. The recent affordability of drone technologies is because of the mass production enabled by SZ DJI technology and other manufacturers of drones in China (Khanna, 2018). However, when manufacturers need solutions that are tailored to sense, move or transform capabilities, the price of drones, consulting and infrastructure will increase significantly.

5.1 Current state of drones in manufacturing

The following observation by a drone vendor serves to illustrate the current state of drones in manufacturing:

Everybody wants a flying Swiss-army knife. But drones aren’t capable of doing everything. It’s really about trying to give the customer real expectations and tailored capabilities; it’s about trying to figure out their primary goals and what they are trying to accomplish with the data.

In 2018, there were few established applications of drones in manufacturing. Many companies are now experimenting with the use of drones in different applications, and a few manufacturers have already begun to use drone applications in warehouse operations and inspection tasks. Nevertheless, there is a significant potential for further drone applications. As drone technology continues to develop during the next 5–10 years, we expect a range of new use cases to emerge across many manufacturing industries. Figure 3 provides a summary of current drone applications used in manufacturing industries.

The majority of the current applications of drones are the “see” and “sense” types. The first drone experiments by most manufacturers involve off-the-shelf commercial drones with high-definition cameras for photos and videos. Many manufacturers employ at least one drone enthusiast who brings his or her interest and expertise to producing aerial photos and videos of the facilities. These applications are inexpensive and simple, and they do not require specialized drone technology or consultation. Learning-by-doing with inexpensive drones helps accumulate knowledge and enables incremental innovation (Bourke and Roper, 2016; Sohal et al., 2006). For manufacturers with large facilities, tanks, hazardous areas, cranes, conveyors or high machines that require regular inspections, the next stop is to consider whether drones could replace manual inspections. In many cases, drones are an economical alternative to traditional inspections. These “see” capabilities could be enhanced to “sense” capabilities by integrating advanced sensors and software. Standard video cameras could be replaced by thermal cameras to detect heat loss from machines and buildings. Gas-sniffing sensors could be used to detect gas leaks. Laser or ultrasound sensors could be used to conduct non-destructive testing in hard-to-reach areas. Barcode or RFID readers could be used to identify objects on high shelves. LiDAR scanners could be used in volume measurement and the indoor 3D mapping of factories, which could help in designing layouts and in factory redesign projects. There are examples of all these drone applications in current manufacturing practices.

It is harder to improve the physical capabilities of drones than it is to improve their analytical capabilities because of their physical limitations, especially their payload and battery capacity restrictions. In “move” or “transform” tasks, which require high physical capabilities, drones are typically inferior to tools, sensors or cameras that are mounted to grounded infrastructure, AGVs, cranes, walls or ceilings. “Move” operations, such as intra-logistics and part delivery applications, are rare, and their use is practical only in small, light, and urgent “emergency” deliveries. A non-manufacturing outdoor example is the delivery of blood samples from hospitals to test laboratories, which already is in daily operation for example in Lugano and Zurich in Switzerland (Müller, 2018). Indoor applications are more difficult to justify. Even in the case of missing parts on an assembly line (see quote in section 4.2); the better option is to remove the root cause of the error. “Transform” operations are rare not only in manufacturing but also in all industries. Some applications are in the experimental stage in the oil and gas industry and the construction and infrastructure industries. In manufacturing, remote maintenance operations and corrosion protection may be promising business cases in nuclear power plants, metal smelting plants, shipyards, petrochemical plants and other large process industry plants. However, the risk of explosion remains a technical hurdle for the full adoption of drones in these contexts.

5.2 A research agenda

Our insights suggest that by 2025, drones will be applied in many manufacturing plants. This should provide a rich opportunity for empirical research in this area. In the support of an efficient development of a literature on drones in manufacturing, we call for three broad streams of research: operations management and industrial engineering issues; technology development and customization for manufacturing; and socio-cultural and behavioral issues.

We call for conducting both descriptive-oriented and solution-oriented research on the applications of drones in manufacturing industries (see, van Aken, 2005). Operations management scholars could extend the literature on the evaluation, implementation and measuring effects of adopting drones in manufacturing. As a starting point in particular, scholars can build frameworks for supporting “make-or-buy” decision-making. Industrial engineering scholars could study specific-use cases and develop design guidelines for different applications. The typology presented in this paper could help focus attention on these areas on research. Because the majority of applications in manufacturing will continue to be of the “see” or “sense” types, we suggest focusing on “sense” as the most promising type of application. Because this field of research is in the early stage, we suggest that qualitative empirical methods should be used to advance the research on drones in manufacturing. Expert interviews, which were used in the present study, are a promising method for explorative research. Action research and design science research can support the development of solution-oriented theories and practical artifacts. Modeling and simulation can help explore and evaluate opportunities in virtual environments. As the number of cases in industry continues to increase, case research and survey research will allow for the exploration of the tangible benefits and challenges of the implementation of drones in manufacturing. The theoretical literature on AMT is a valuable starting point. For instance, reviewing AMT literature shows the potential of survey research to determine the industries that are more likely to benefit from each type of applications, as well as elaborating on the contingency factors (such as proximity to know-how, size of the companies and organizational culture) that influence drone adoption in manufacturing companies.

There is also much work to be done in the engineering sciences and in product development regarding the application of drones in manufacturing. The further development of drone technology, such as the capability of carrying heavier loads, conducting long missions (e.g. drones with hydro engines) and improving the technology of BVLOS with advanced object recognition and object avoidance algorithms, will increase the number of future applications. Research could also help improve battery life, communication and control processes, safety systems, the availability of sensors and other technological aspects. Wu et al. (2015) suggested that cloud-based design and manufacturing could increase the productivity of the design, prototype and production of future drones, thus enabling faster reactions to market needs.

The most promising technological research in terms of manufacturing applications is the development of automatic drones. Replacing manual work by a piloted drone produces only marginal benefits. Replacing manual work by automatic drones is a much better business proposition. Automatic drones require complementary technologies, such as ground stations, multiple sensors for object recognition and avoidance, and algorithms that control flight. The next step after automatic flight is autonomous flight by artificially intelligent drones that make decisions in changing environments (Floreano and Wood, 2015). One interviewee from a leading provider of drone-based inspection stated, “We are long way away from autonomy, but we are close to automation.” Autonomous micro aerial vehicles is another research area with promising potentials in manufacturing operations (Kumar and Michael, 2012). Furthermore, the “swarming drone” technology could offer unprecedented opportunities to scale drone applications for inventory management and material handling (Khosiawan, Khalfay and Nielsen, 2018). However, this technology would require robust wireless communication, three-dimensional trajectory data, precise flight control and scheduling task execution (Khosiawan and Nielsen, 2016; Khosiawan, Park, Moon, Nilakantan and Nielsen, 2018), all of which is worthy of future research.

Drones differ from many other AMTs because they are negatively portrayed by media. In addition, drones exhibit animal-like behavior, they are noisy, and they can be hard to see until they are close to their target. People know that drones collect data and can potentially film them while they are working, which poses serious questions about personal data protection rights. In addition, drones and robotics in general evoke the fear that people will lose their jobs to machines (Stewart, 2015). In short, drones involve a trust problem that is more serious than that involving many other AMTs. To establish trust for drone applications in manufacturing, past studies on AMT advice managers to develop an innovation-supportive culture that supports experiments with new technologies (Khazanchi et al., 2007). Because drones are quite different from other “grounded” AMTs, socio-cultural and behavioral aspects of drone implementation represents a particularly promising research area. Such research on behavioral aspects could be based on ethnographic, field experiment, interview or survey methodologies.

5.3 Implications for practice

Drones are a new form of AMT that will be applied in many manufacturing industries, especially in large, technology-intensive facilities in process industries. The overview of current use cases shown in Figure 3 could provide manufacturers with a perspective on what is possible today. An important point is that current drone applications are mainly “see” and “sense” types of applications.

Where should manufacturers start? As the arrows shown in Figure 3 indicate, manufacturers could start with simple experiments related to the “see” capabilities. From there they could move to “sense” applications, “move” applications or both. However, the transition to “move” applications is currently the most challenging. By following this advice, manufacturers could start running experiments with off-the-shelf drone technologies. Such actions could foster learning and champion drone technology through familiarization and promotion (Dimnik and Johnston, 1993; Kolb, 1976). That would help in discerning opportunities and challenges, as well as justifying investment (Boyer, 1999; Kolb, 1976). As suggested in previous work on AMT, learning from the experiences of other manufacturing companies can help managers avoid common mistakes and assist them during the planning phase of a drone program (Sohal, 1996). Manufacturers that have gained experience in using drones in “sense” or “move” tasks could consider further integrating technology to make their drones capable of performing “transform” tasks. Evidence from the adoption of other AMTs imply that such integration needs financial and strategic justification, readiness for organizational change, investment in infrastructure and support from top management (see, Small, 2007; Gouvea da Costa and Pinheiro de Lima, 2008; Gouvea da Costa and Pinheiro de Lima, 2008; Dean et al., 1992; Zammuto and O’Connor, 1992; Boyer et al., 1997; Percival and Cozzarin, 2009; Bessant, 1994).