ASABE Logo

Article Request Page ASABE Journal Article

Identification of Advantages and Limitations of Current Risk Assessment and Hazard Analysis Methods when Applied on Autonomous Agricultural Machineries

Guy R. Aby1, Salah F. Issa1,*, John F. Reid2, Cheryl Beseler3, John M. Shutske4

Published in Journal of Agricultural Safety and Health 30(2): 35-52 (doi: 10.13031/jash.15873). Copyright 2024 American Society of Agricultural and Biological Engineers.

1    Agricultural & Biological Engineering, University of Illinois at Urbana-Champaign, Urbana, Illinois, USA.

2    Computer Science/Ag and Bio Eng/ECE, University of Illinois at Urbana-Champaign, Urbana, Illinois, USA.

3    Department of Environmental, Agricultural and Occupational Health, University of Nebraska Medical Center, Omaha, Nebraska, USA.

4    Biological Systems Engineering, University of Wisconsin, Madison, Wisconsin, USA.

*    Correspondence: salah01@illinois.edu

The authors have paid for open access for this article. This work is licensed under a Creative Commons Attribution NonCommercial NoDerivatives 4.0 International License https://creative commons.org/licenses/by nc nd/4.0/

Submitted for review on 28 October 2023 as manuscript number JASH 15873; approved for publication as a Research Article and as part of the Safety for Emerging Robotics & Autonomous Agriculture Collection by Associate Editor Dr. Farzaneh Khorsandi and Community Editor Dr. Michael Pate of the Ergonomics, Safety, & Health Community of ASABE on 21 March 2024.

Citation: Aby, G. R., Issa, S. F., Reid, J. F., Beseler, C., & Shutske, J. M. (2024). Identification of advantages and limitations of current risk assessment and hazard analysis methods when applied on autonomous agricultural machineries. J. Agric. Saf. Health, 30(2), 35-52. https://doi.org/10.13031/jash.15873

Highlights

Abstract. In the last ten years, the development of automated agricultural machinery has seen noteworthy advancements. Nevertheless, the successful commercialization of these technologies depends critically on their ability to operate safely. This study evaluated the advantages and limitations of current risk assessment and hazard analysis methods currently used to ensure the safety of autonomous agricultural machines. An online survey containing 18 questions was distributed to 711 participants identified as potential individuals who are currently working or have worked on autonomous agricultural machines to determine the type and frequency of risk assessment and hazard analysis methods applied on autonomous agricultural machines, examine the advantages and limitations of each method, and investigate the perceived effectiveness of each method. Frequency analysis was used to determine the most and least utilized risk assessment and hazard analysis methods. The advantages and limitations of each risk assessment and hazard analysis approach were compared. Descriptive statistics (counts, means, medians, percent) and frequency analysis of the variables were used. The three main types of risk assessment and hazard analysis techniques applied to autonomous agricultural machines. The methods are (a) Informal Group Analysis (e.g., Brainstorming), (b) Hazard Analysis and Risk Assessment (HARA), and (c) Failure Mode and Effects Analysis (FMEA). Replicability is perceived as the main advantage of FMEA and HARA, while cost-effectiveness is the main advantage of Informal Group Analysis. The need to have pre-existing data of the autonomous agricultural machine at hand to be able to perform risk assessment and subjectivity are the main limitations of FMEA, HARA, and Informal Group Analysis dealing with novel and revolutionary autonomous agricultural machines. Industry experts do not believe that the risk assessment and hazard analysis procedures now used are reliable and efficient enough to guarantee the safety of autonomous agricultural tractors. This study reveals important information about the current state of risk assessment and hazard analysis methods in the context of autonomous agricultural machinery. This knowledge can inform future research, policy development, and industry practices to ensure the safety of autonomous agricultural machines.

Keywords.Agricultural machine, Autonomous, FMEA, HARA, Hazard analysis, Informal Group Analysis, Risk assessment, Safety, Survey.

The agricultural landscape is undergoing a profound transformation. The utilization of cutting-edge automation, artificial intelligence, and robotics in agriculture has given rise to a new generation of production agriculture practices. Autonomous agricultural machinery, equipped with a variety of sensors, processors, and actuators, has the capability to plow, sow, cultivate, and harvest crops with unprecedented precision and efficiency (Oliveira et al., 2021; Shamshiri et al., 2018; Lytridis et al., 2021). As this shift becomes more prominent, ensuring the safety of these autonomous machinery remains crucial (Aby et al., 2024).

The significance of ensuring the safety of these advanced machines goes beyond operational efficiency. It extends into the domains of ethics, economics, ecological preservation, and operator/public safety. As these machines integrate into the agricultural landscape, they interact with humans, livestock, and ecosystems. The imperative to ensure safety is important from two main perspectives. First, it is a moral obligation to prevent harm to human operators, bystanders, and animals who share the fields with these autonomous machines. Second, safety is an essential foundation for the widespread adoption and acceptance of autonomous agricultural technologies (Rial-Lovera, 2018). A single unfortunate incident can erode trust in these innovations, stalling progress and hindering the realization of their transformative potential. This is currently observed in the State of California, where the Division of Occupational Safety and Health (DOSH) board, better known as Cal/OSHA, voted to restrict autonomous tractors (Woelfel, 2022).

Recently, a comprehensive literature review conducted by Aby and Issa (2023) revealed that previous research efforts to ensure the safety of autonomous agricultural machines have focused on three main areas, including (1) environmental perception, (2) risk assessment and hazard analysis, and (3) human factors and ergonomics. However, little effort has been allocated towards the area of risk assessment and risk mitigation. From a safety engineering perspective, safety evaluation is performed through risk assessment and hazard analysis methods such as Fault Tree Analysis (FTA), Failure Mode and Effects Analysis (FMEA), Human Factors Analysis and Classification System (HFACS), and Hazard Analysis and Risk Assessment (HARA). Hazards are classified into three categories: inherent properties or characteristics of the product or its components/subsystems, human failure, and environmental stresses. The hazard analysis is required to identify such hazards, determine which hazards can be eliminated through the design process, and determine the best approach for reducing remaining hazards to an acceptable degree (Murphy, 1999).

Risk assessment and hazard analysis techniques have been used in diverse industries and technologies for several years (Tucan et al., 2020; Rathour et al., 2020). However, a recent survey conducted by Shutske et al. (2023) has shown that most industry professionals use a lower number of risk assessment and hazard analysis techniques with autonomous agricultural machinery than they would when analyzing non-automated agricultural machinery. One possible reason is due to the fact that autonomy is relatively new, standards and risk assessment methods take time to evolve, and there are just “fewer” methods available, and those that are useful have issues with a lack of data. However, overall, it is unclear why professionals in the field no longer feel comfortable employing these methods when dealing with automated agricultural machinery. Finding all the reasons why industry professionals no longer feel confident using these risk analysis methods would reveal the specific weaknesses and strengths of these methods. Subsequently, this will help engineers to modify and/or design well-suited and efficient risk assessment analysis tools for autonomous agricultural machinery so that safety concerns are lessened and hopefully will present less of a barrier to adoption for the industry and to the stakeholders who might benefit from new advances in autonomous technology.

Therefore, the objective of this study is to build upon the recent study conducted by Shutske et al. (2023) to identify the types and frequency of risk assessment and hazard analysis techniques used on autonomous agricultural equipment, to investigate the methods' perceived effectiveness, and to examine the advantages and limitations of each technique.

Materials and Methods

A survey was conducted to provide explanatory insight into the type and frequency of risk assessment and hazard analysis methods applied on autonomous agricultural machines, the perceived effectiveness of each method, as well as their respective strengths and weaknesses.

Survey Instruments

The survey was developed using the Qualtrics XM software package, which allows users to develop questions, program logic flows, track distribution, and analyze collected data for desktop computer or mobile response. Survey recipients were presented with a clear definition of “automated,” “autonomous,” and “remote operator.” The definition was adopted from the International Organization for Standardization's ISO 18497 (2016). The survey consisted of 18 questions. A total of six experts in the field have revised the questionnaires to ensure their content validity. All questions were coded within the survey software as mandatory, requiring participants to answer each question before proceeding to the next question. Input validation within Qualtrics was used to ensure that a response was given to each question for the respondent to progress within the survey. The survey's first question asked participants to confirm they had read and understood the consent form and to certify that they were 18 years of age or older. This is the first criteria that needs to be met by participants to be able to take the survey. The second question of the survey asked participants if they have ever or currently worked with the design, development, testing, application, research, or development of autonomous agricultural machinery. Simple Boolean logic was used to filter the respondents that indicated they did not meet these criteria, directing them to the end of the survey. This filtering process was used to ensure that only individuals who have worked with autonomous agricultural machinery would provide data.

Participants

The American Society of Agricultural and Biological Engineers (ASABE) committee name titles were used to determine possible survey participants. A total of 252 committee names were identified and examined. Among the 252 committee title names, to identify only the committees that are likely to include a scope and activities relevant to automated agricultural machinery, the committee title names with one or a combination of the following keywords were extracted: Ergonomics; Safety; Health; Machine; Robotics; Mechatronics; Instrumentation; Machinery; Transport; Tractors; Electronics; ROPS; Equipment; Automation; Aerial; precision. Overall, 31 committee title names were obtained with 1746 members (table A1). Members' names, positions held, organizations they belong to, and email addresses were exported to a Microsoft Excel file. The CO-PI of this study and duplicate names that resulted from people serving on several committees were deleted. As a result, 711 members were obtained and considered in this study. The study recruitment protocol and survey tool were submitted and approved by the institutional review board (IRB) as required by university and federal policy for any study involving human subjects.

Consent

Participating in this study presented a possible risk to respondents if a company was sensitive about disclosing the techniques and methods used to assess risk in relation to highly automated agricultural machinery. To mitigate this potential risk, the aggregated list of potential survey respondents was stored separately from other study data under two-factor authentication required for accessing the data. Additionally, participants were emailed an anonymous link to access the Qualtrics survey, which prevented any ability to connect individuals to their survey responses. Participants were informed about all aspects of the study during the consent process and had the option to withdraw from any activity that made them uncomfortable at any point. Participants were asked to complete a second survey to provide their name and email address and receive $5 incentives while maintaining the anonymity of their survey responses, as well as a fact sheet that summarized the results of the survey. The information of those participants was stored under two-factor authentication, separate from any other data, until the compensation was distributed, then the information was destroyed.

Survey Delivery

To motivate participants to respond, the identified individuals were emailed a total of three times with a link to the survey and additional information as recommended by Dillman et al. (2011). The initial request included a cover letter and IRB-required consent information. Two reminder emails were sent: one after the first full week and one three weeks post-initial contact, respectively. There was a $5 incentive for completing the survey. Participants were asked to complete a second survey to receive their $5 incentives while maintaining the anonymity of their survey responses.

Data Analysis

The main objective of this study was to determine how frequently engineers apply various risk assessment and hazard analysis techniques while working with autonomous agricultural machinery, as well as their individual perceived benefits and limits. The survey contained 19 questions that were grouped into four main categories:

The most and least often used risk assessment and hazard analysis techniques were identified using frequency analysis. The advantages and limitations of each risk assessment and hazard analysis approach were compared and contrasted. To look for patterns, descriptive statistics (counts, means, medians, percent) and frequency analysis of the variables were used.

Results

Overall, 711 surveys were distributed to individuals identified as potentially working with autonomous agricultural machinery. Of the distributed surveys, 120 were started. This represented an overall response (opening the survey) of 16.88%. Of the 120 responses, 68 (56.67%) indicated that they have at some point worked with the design, development, testing, application, or research and development of autonomous agricultural machinery, while 52 (43.33%) indicated that they have never done work related to autonomous agricultural machinery. Of the 68 responses, 16 (23.53%) did not move forward to answer the questions, 12 (17.65%) answered the first 3 questions but did not complete the survey questionnaires, and 40 (58.82%) have answered all the survey questions. The findings in figure 1 are based on the responses of 52 participants, since 52 people have answered the first three questions. Afterwards, the remaining results focus on the 40 participants who have completed all the survey questions.

Initially, participants were asked to specify the type of risk assessment and hazards analysis methods they use with autonomous agricultural machines, as well as the frequency with which they use each method. Eight risk assessment and hazards analysis methods were presented as choices in the survey, with the additional option of writing in a non-listed risk assessment and hazards analysis methods. These methods were: Failure Mode and Effects Analysis (FMEA), Hazard Analysis and Risk Assessment (HARA), Fault Tree Analysis (FTA), Hazard and Operability Analysis (HAZOP), Systems Theoretic Process Analysis (STPA), Monte Carlo Analysis, Risk Analysis Software, and Informal Group Analysis. The results show that Informal Group Analysis, FMEA, and HARA are the most used risk assessment and hazards analysis techniques, while STPA, Monte Carlo Analysis, FTA, and HAZOP are the least used techniques. The results are presented in figure 1.

Follow up questions were asked to (1) select the method (only one choice) individuals have used the most at their current company and (2) assess the effectiveness of that method. The results are presented in figures 2 and 3. The results show that Informal Group Analysis was the most used method with 13 selections (33%), followed by 10 selections (25%) for HARA, 8 selections (20%) for FMEA, and 6 selections (15%) for the Monte Carlo Analysis. The FTA, HAZOP, and STPA had a single (3%) selection, respectively. Moreover, regarding the perceived effectiveness of the risk assessment and hazards analysis methods, when looking at all the responses together regardless of the type of methods, “somewhat effective” accounted for 27 (68%) of responses, then 8 (20%) for “neutral,” 3 (8%) for very effective, and 1 (3%) for “somewhat ineffective” and “very ineffective,” respectively. Additionally, when looking at the perceived effectiveness level for each risk assessment and hazard analysis individually, both the Monte Carlo Analysis and HARA have about 70% for somewhat effective and about 30% for neutral. On the other hand, FMEA has 88% for being somewhat effective and 13% for being very effective. Finally, the Informal Group Analysis has 54% for somewhat effective, 23% for neutral, and 8% for very effective, somewhat ineffective, and very ineffective, respectively. The results are shown in figure 4.

Figure 1. Frequency with which different risk assessment and hazard analysis methods have been applied on autonomous agricultural machines.

Another question was posed to explain the "advantages" and "limitations" of each approach that was chosen as the one that was used the most. Tables 1 and 2 present the findings. The results for FTA, HAZOP, and STPA are not taken into account in tables 1 and 3 due to the smaller percentages of participants (3% each) who chose those methods as their most used technique. Replicability and the software's wide availability (with a selection of 100% and 25%, respectively) are the main advantages of FMEA. The main advantages of HARA are that it is replicable and fast, with 73% and 13% of selection, respectively. The main advantages of Informal Group Analysis, selected by 36% and 32% of respondents, respectively, are its speed and cost-effectiveness. Last but not least, similar to HARA, the main advantages of the Monte Carlo Analysis are replicability and time efficiency, with 33% and 33% of selection, respectively.

Figure 2. Most Commonly Used Risk Assessment and Hazard Analysis Techniques at Respondents' Companies.
Figure 3. Percentages of each effectiveness level of all the risk assessment and hazard analysis techniques.

Additionally, participants were given a blank area where they could put "other" benefits of the risk assessment approach they had chosen. Regarding the FMEA, one respondent said that "most risk assessment people are familiar with it" and another said that "FMEA is very easy to iterate design concepts with." One respondent described HARA as having "high resolution with the ability to cover very large differences in input value," while another stated that "the method is thorough." The following three justifications were given as to why Informal Group Analysis is most frequently used: "the best fit when other methods don't work," "readily available," and "the formal process is too constricting to the way that we work in problem identification."

The results in table 3 show that “the need to have pre-existing data” and “subjectivity” are the top 2 limitations of FMEA, HARA, and Informal Group Analysis. On the other hand, “time consuming” and “cost” are the top two limitations of the Monte Carlo Analysis. Similarly, participants were provided with a blank box allowing them to write “other” limitations of their selected risk assessment method. Overall, only one respondent provided comment on HARA. His comments include “limited direction on use of process,” and “low acceptance by third parties.”

Figure 4. Percentages of each effectiveness level for each the risk assessment and hazard analysis techniques.
Table 1. Frequency of different advantages of each risk assessment and hazard analysis technique.
Type of
Advantages		FMEAHARAInformal Group
Analysis		Monte Carlo
Analysis
Readily available software25%7%9%17%
Replicable100%73%23%33%
Not time consuming8%13%32%33%
Not expensive17%7%36%17%

Table 2. Frequency of different limitations of each risk assessment and hazard analysis technique.
Type of
Limitations		FMEAHARAInformal Group
Analysis		Monte Carlo
Analysis
Requires pre-existing data33%23%30%10%
Subjectivity22%23%30%0%
Time consuming11%23%3%30%
Does not consider artificial intelligence17%8%13%20%
Expensive6%0%0%30%
Undefined requirements (next step)11%23%23%10%

Furthermore, to gain a deeper understanding of the rationale behind the choice of their primary risk assessment technique, participants were prompted to select the specific reasons for favoring that particular method. The results are presented in table 3.

"Best method" and "proven track evidence" account for 23%, 54%, 55%, and 27% of the selection for FMEA and Informal Group Analysis, respectively. "Company requirement" and "best method" account for 58% and 25% of the selections in HARA, respectively. Finally, "best method" and "cost effective" accounted for 44% and 33% of the selections for the Monte Carlo Analysis, respectively.

Table 3. Frequency of different reasons why each risk assessment and hazard analysis technique is used the most.
Type of
Reasons		FMEAHARAInformal Group
Analysis

    Monte Carlo

Analysis
Best method23%25%55%44%
Proven track evidence54%17%27%22%
Company requirement15%58%0%0%
Cost effective8%0%18%33%

Participants were asked about the ways in which risk assessment processes and tools could be improved and made more effective for their use with autonomous agricultural machines. The ability to use tools that fully consider and account for the presence of AI in their functions was cited most frequently, along with risk assessment tools being available in a discrete “checklist” format to aid the evaluator(s) doing risk assessment. Other areas where there was cited potential for improvement included tools or information to assist in interpretation and actions required that resulted from risk assessment efforts, as well as the availability of apps or other software that would serve as decision support or guidance tools to walk persons or teams through a robust risk assessment process” (table 4). In the “other (please, if possible, explain)” box area, one respondent expressed a desire to have “well-defined operational design domains," another mentioned “clear examples of how the process is applied to complex systems,” and the last one wrote “a clear description of the system pipeline, including parts and functionalities.”

Table 4. Frequency of different potential improvement considerations required in current risk assessment and hazards analysis techniques when applied on autonomous agricultural machines.
Improvements in Risk
Assessment Analysis Methods		NFrequency
(%)
Parameters to consider Artificial Intelligence (AI)2973%
Established process or checklist2973%
Clear description and interpretation of result1948%
Software or Application (Apps)1640%

Table 5. Frequency of different factors considered during the risk assessment and hazard analysis process.
FactorsNFrequency
(%)
Safety40100%
Functional reliability3895%
Training and education3587.5%
Machine capabilities and limitations3177.5%
Regulatory compliance2972.5%
Environmental impact2562.5%
Emergency response mechanisms2152.5%
Cybersecurity2050%

Table 5 shows the frequency with which different factors have been considered during the risk assessment process for autonomous agricultural machines. “Safety” refers to evaluating potential risks to human operators and interaction with workers, bystanders, and animals in the vicinity of the machines. Assessing the likelihood and severity of accidents, such as collisions, entanglements, or crushing incidents, “Functional reliability” refers to examining the reliability and robustness of the autonomous systems, including sensors, actuators, and communication modules. Consider potential failures or malfunctions that could lead to accidents or operational disruptions. “Cybersecurity” refers to analyzing the vulnerability of the autonomous agricultural machines to cyber threats, such as unauthorized access, data breaches, or system manipulation. Assess the potential impact on safety, privacy, and the integrity of data and operations. “Environmental impact” refers to assessing the potential environmental consequences of autonomous agricultural machines. Consider factors such as soil compaction, uneven terrain (slope, hill), chemical usage, wind, and emissions. “Regulatory compliance” refers to ensuring that the autonomous agricultural machines comply with relevant regulations and standards, including safety standards, emissions regulations, and data privacy laws. Consider any legal or regulatory requirements specific to the region or industry. “Training and education” refers to evaluating training requirements for operators and maintenance personnel. Assess the level of knowledge and skills needed to operate, program, and maintain autonomous machines safely and efficiently. “Machine capabilities and limitations” refers to evaluating the abilities and constraints of the machinery at hand. This assessment is crucial to understanding how well the machine system can perform its intended tasks and where potential weaknesses or limitations may exist. “Emergency response mechanisms” refers to the systems and procedures in place to address and mitigate potential emergencies or unexpected situations.

Table 6 presents the titles and roles of individuals who have employed risk assessment and hazard analysis techniques for autonomous agricultural machines, and table 7 highlights the specific types of autonomous agricultural machines upon which these techniques have been applied.

Table 6. Frequency of participant roles within their company of current employment.
Roles
(Job Titles)		NFrequency
(%)
Agricultural Engineer2358%
Product safety1538%
Product compliance718%
Product reliability615%
Software Engineer615%
Electrical Engineer615%
Mechanical Engineer410%
Automotive Engineer25%
Environmental and Civil Engineer | Robotics researcher and group manager | Principal Systems Engineer | Research and education | Educator1 (each)3% (each)

Furthermore, three additional questions explored: (1) how many years respondents have worked on autonomous agricultural machinery; (2) how many employees their respective company has; and (3) how long the company they are currently working for has been around. The results are presented in figures 5, 6, and 7.

Regarding the number of years of experience, 57.5% of respondents have 0–10 years of experience, 20% have 11–20 years, and 22.5% have 21+ years of experience. A total of 57.5% of survey participants have 0 to 10 years of experience, while 42.5% have 11 or more years of expertise.

For participants with 0–10 years of experience, HARA is the most frequently used method, 7 selections (30%), followed by 6 selections (26%) for the Informal Group Analysis, 4 selections (17%) for FMEA and Monte Carlo Analysis, respectively, and lastly, 1 selection (4%) for FTA and STPA, respectively.

Table 7. Frequency of the type of autonomous agricultural machines participant have worked on.
Type of Autonomous
Agricultural Machines		NFrequency
(%)
Driverless tractor2153%
Driverless harvester1538%
Drone1333%
Ground sprayer1333%
Scouting923%
Phenotyping820%
Weeder513%
Carrying worker38%
Mower25%
Pruning | Snow removal | Seeder | Spreader | Forage wagon | Robotic milking | Animal Shade and Feed Delivery | Sugar cane logistics | Grain handling | Tillage robot1 (each)3% (each)
Figure 5. Distribution of Participants by Years of Experience
Figure 6. Percentage Breakdown of Company Sizes Among Participants

For participants with 11+ years of experience, Informal Group Analysis is the most frequently used method, with 7 selections (41%), followed by 4 selections (24%) for FMEA, 3 selections (18%) for HARA, 2 selections (12%) for Monte Carlo Analysis, and lastly, 1 selection (6%) for HAZOP.

Overall, for participants with 0–10 years of experience, HARA and Informal Group Analysis are the most used risk assessment methods, while for participants with 11+ years of experience, Informal Group Analysis and FMEA are the most used risk assessment methods. Regardless of the number of years of experience of the participants, Informal Group Analysis (e.g., brainstorming) is the most used method.

As far as company size, 52.5% of respondents fall within companies that have less than 1000 employees, 27.5% fall within 1001–2000 employees, and lastly, 20% fall within 20000+ employees. Among the respondents, 52.5% belong to companies with fewer than 1000 employees, while 47.5% are associated with companies employing 1001 or more individuals.

For companies with fewer than 1000 employees, Informal Group Analysis is the most frequently used method with 7 selections (33%), followed by Monte Carlo Analysis with 5 selections (25%), FMEA with 4 selections (19%), HARA with 3 selections (14%), and lastly, FTA and STPA, each with 1 selection (5%).

For companies that have 1001+ employees, HARA is the most frequently used method, 7 selections (37%), followed by 5 selections (32%) for Informal Group Analysis, 4 selections (21%) for FMEA, lastly 1 selection (5%) for Monte Carlo Analysis and HAZOP, respectively.

Overall, Informal Group Analysis and Monte Carlo Analysis are the most used risk assessment methods in companies with fewer than 1000 employees, while HARA and Informal Group Analysis are preferred in companies with 1001 or more employees.

Regarding how long the company has been around, 22.5% of respondents fall within companies that have less than 10 years, only 5% fall within 10–20 years, and 72.5% fall within 20+ years.

Figure 7. Percentage Distribution of Company Ages Among Participants

This study aimed to explore the type and frequency of risk assessment and hazard analysis methods applied to autonomous agricultural machines, along with an assessment of their perceived effectiveness, advantages, and limitations. Furthermore, the study explored the motivations behind the use of these risk assessment methods and identified areas for improvement to enhance the safety of autonomous agricultural machines.

While engineers employ a range of risk assessment and hazard analysis techniques to ensure the safety of autonomous agricultural machines, the results of this study show that Informal Group Analysis (Brainstorming), HARA, and FMEA are the most utilized methods. Among these, Informal Group Analysis is the most frequently used method, with 13 selections (33%), followed by 10 selections (25%) for HARA, 8 selections (20%) for FMEA.

Informal Group Analysis remains the top choice for several reasons. First and foremost, brainstorming is a relatively simple, inexpensive, and accessible method that does not require extensive training or specialized tools. The result of this survey shows that the key advantages of Informal Group Analysis are that it is not time consuming and not expensive, with 36% and 32% of responses, respectively. This makes it a practical choice for involving various stakeholders in the safety assessment process, including farmers, machine operators, and other relevant parties. Moreover, Informal Group Analysis sessions involve gathering a group of individuals with diverse backgrounds and expertise. This diversity can lead to a rapid generation and a wide range of ideas, insights, and perspectives on potential safety risks that may not be evident to a single individual or through formalized methods. As a result, safety engineers can quickly identify potential risks associated with autonomous agricultural machines, making it a valuable method when time is limited or there is a need for timely risk assessment.

Both HARA and FMEA also have several advantages. These methods have systematic processes that help identify hazards and failure modes early in the design process and are replicable. Indeed, this study's results unequivocally demonstrate that replicability stands out as the most advantageous factor for both FMEA (100% of the responses) and HARA, which received 73% of the responses. These methods can be applied early in the development process of autonomous agricultural machines, allowing engineers to address safety concerns at the design stage. This is crucial for ensuring that safety features are integrated into the system from the beginning, reducing the need for costly retrofitting or modifications later.

Moreover, HARA and FMEA require structured documentation of the identified hazards, their causes, consequences, and mitigation measures. This documentation ensures transparency and accountability throughout the development process, making it easier to track safety improvements and communicate them to stakeholders. Additionally, HARA and FMEA provide mechanisms for prioritizing risks and failure modes based on severity, likelihood, and detectability. This helps engineering teams focus their resources on addressing the most critical safety concerns, thereby optimizing safety efforts. Finally, HARA and FMEA are iterative processes that encourage continuous improvement. As autonomous agricultural machines evolve, safety engineers can revisit and update their risk assessments to account for changes in technology, operational conditions, and usage patterns.

The need to have pre-existing data and subjectivity are the top 2 limitations of FMEA, HARA, and Informal Group Analysis when applied to autonomous agricultural machines. Autonomous agricultural machines operate in dynamic and unpredictable environments, which can vary greatly depending on factors like weather, soil conditions, and crop types. To perform a thorough risk assessment, engineers typically need historical data and information such as the specific machine's performance, the agricultural environment, potential failure modes, potential applications or misuses of machines, and previous accidents. Therefore, limited pre-existing data can hinder the accuracy and completeness of the analysis. For example, without sufficient data, it may be difficult to estimate the likelihood of specific failures occurring, such as sensor failures in dusty or muddy conditions or the impact of software glitches on navigation in different terrains. It appears that there are different approaches to dealing with the problem of the lack of available prior data. In recent research by Muller et al. (2022), the authors employed digital twin to supply the necessary data in order to overcome the shortage of data for conducting risk assessment analysis.

Risk assessments often involve subjective judgments and assumptions made by experts or team members involved in the analysis process. In the case of autonomous agricultural machines, subjectivity can introduce bias and inconsistencies into the risk assessment process. Different experts may have varying opinions on the severity of certain failure modes or the likelihood of specific events occurring. Therefore, subjectivity can also lead to disagreements and disputes among team members, which can hinder the effectiveness of risk management efforts. For example, one team member might believe that a particular sensor failure is a minor issue, while another might consider it a critical safety concern, depending on their individual perspectives and experiences.

Overall, only 8% of participants believe that the risk assessment and hazard method they use is very effective in ensuring the safety of autonomous agricultural machines. The remaining 82% of participants, on the other hand, perceive their chosen risk assessment and hazard methods as falling within the spectrum of somewhat effective, neutral, somewhat ineffective, or very ineffective. These findings suggest that current risk assessment and hazard analysis techniques are not sufficiently equipped to ensure the safety of autonomous agricultural machines. This observation is not unexpected, given the complex nature of these machines, which feature advanced technologies such as sensors, machine learning algorithms, and complex control systems. These systems can interact in intricate ways, making it challenging to predict and identify all possible failure modes and risks through traditional methods like HARA. Furthermore, as mentioned above, traditional risk assessment methods often rely on historical data to quantify risks. However, autonomous agricultural machines are relatively new technologies, and there may be limited historical data available for such risk analysis.

Additionally, autonomous systems often incorporate machine learning algorithms, which can be challenging to analyze using traditional deterministic risk assessment techniques. The behavior of machine learning models can be non-deterministic, making it difficult to identify and assess potential hazards. Autonomous systems can learn and adapt to changing conditions, which means that their behavior can evolve over time. This adaptability makes it difficult to predict and address all potential risks through static risk assessment techniques. Agricultural environments are dynamic, with variations in terrain, weather, and crop conditions. Traditional risk assessments may not fully account for these real-world variabilities when assessing safety risks.

Participants were asked to select and/or propose potential enhancements to risk assessment and analysis methods, aiming to strengthen the safety of autonomous agricultural machines. The majority of participants (73%) chose "parameters to consider AI" and "established processes or checklists" as their preferred options. While these improvements in conventional risk assessment methods hold value in safety analysis, the complexity and dynamic nature of autonomous agricultural environments necessitate a more comprehensive and adaptable approach. Novel methodologies are imperative, ones that can model risk dynamically, taking into account the specific situation—a concept known as situational risk assessment (Muller et al., 2022; Rathour et al., 2020).

The results of this study show that several factors were considered during the risk assessment and hazard analysis processes. Such factors include safety, functional reliability, training and education, machine capabilities and limitations, regulatory compliance, environmental impact, emergency response mechanisms, and cybersecurity. Among these factors, environmental impact, emergency response mechanisms, and cybersecurity had the lowest frequency of considerations with 62.5%, 52.5%, and 50%, respectively.

Environmental impact is crucial and therefore should always be considered during the risk assessment process. A machine that performs an application procedure with fertilizer, livestock manure, pesticides or other product presents a potential environmental risk if an error or malfunction were to occur and lead to a spill, overturn, or application in the wrong area or in certain conditions (like herbicide application in windy conditions). A failure in the field or other operating locations, such as an overturn, would likely lead to a costly downtime incident and/or damage to the machine or other property. Similarly, emergency response mechanisms are also essential. Autonomous agricultural machines, such as self-driving tractors or harvesters, can operate near humans, including farmers and bystanders. In the event of a malfunction or unexpected hazard, having effective emergency response mechanisms can prevent accidents and injuries. Moreover, many regions have specific regulations and standards in place for autonomous agricultural machines.

Implementing emergency response mechanisms is often a legal requirement to ensure compliance with safety and environmental regulations. Lastly, cybersecurity is also critical to ensuring the safety of autonomous agricultural machines. Autonomous machines can be controlled remotely, allowing farmers or operators to monitor and manage them from a distance. If these control systems are not secure, they can be vulnerable to unauthorized access, potentially leading to equipment misuse or damage.

Furthermore, autonomous agricultural machines often rely on sensors and collect vast amounts of data related to crop conditions, soil quality, and machinery performance. Ensuring the cybersecurity of these systems is essential to protecting sensitive data from theft or manipulation. According to research conducted by Drewry et al. (2019), security and privacy were the top concerns among farmers who used early automation and other digital technology.

Correspondence analysis for categorical data, a statistical method explained in Sourial et al. (2010), was conducted to identify relationships among variables to answer three additional research questions: (1) What type of risk assessment method is used with what type of autonomous machinery? (2) What type of engineer is working with what types of automated agricultural machinery? (3) Are certain types of risk assessment methods more common among more experienced engineers than they are among less experienced engineers? Unfortunately, the statistical test did not provide us with better insights, given that we had only 40 participants in this study. Such tests perform well with large sample sizes.

Overall, the efficacy of current risk assessment and hazard analysis methods is contingent upon the availability of pre-existing knowledge about the equipment in question. However, these methods encounter limitations when dealing with novel and revolutionary technologies, especially those lacking pre-existing data, as is the case with autonomous agricultural machines. The advent of such cutting-edge technologies necessitates a paradigm shift in safety evaluation approaches.

Consequently, researchers could focus on developing methodologies that go beyond static assessments and instead embrace dynamic and adaptive frameworks. Such methods would be capable of adjusting to the ever-evolving nature of technological advancements within the agricultural sector. A promising future research project could be the development of real-time situational safety risk assessment methods. By integrating real-time data acquisition and processing capabilities, these advanced assessment methods can offer a proactive and responsive mechanism. This dynamism enables the system to continually analyze and interpret data from the autonomous agricultural machines as they operate in real-world scenarios. By doing so, these methods can identify potential safety risks as they emerge, allowing for timely intervention and mitigation strategies.

Moreover, ISO standard 18497 (2016) contains guidelines for ensuring the safety of agricultural autonomous machinery. However, several important factors to ensure the safety of agricultural autonomous machinery are not considered within the standard. For example, the safety guidelines provided within the standards only apply to “field operations,” not to scenarios or settings in which human operators, service staff, or others might carry out repairs, travel on roadways, or mount/dismount equipment. Injuries sustained during these operations have been recorded and confirmed in previous studies (Shutske et al., 2023; Lee et al., 1996). Furthermore, there are no operations in farmyards or barns, nor on public roads. Moreover, there is no rationale for accommodating maintenance programs, whether executed in the field or in a traditional maintenance shop. However, according to research by Gerberich et al. (1998), injuries with conventional farm machinery frequently happen when performing maintenance, adjustments, and repairs. Consequently, future research efforts could focus on improving the quality of ISO standard 18497 (2016).

Conclusions

The primary goal of this study was to examine the types and frequency with which various risk assessment and hazard analysis methods are used on autonomous agricultural equipment, as well as the perceived efficacy, advantages, and disadvantages of each technique. The key findings are as follows:

  1. The three main types of risk assessment and hazard analysis techniques applied on autonomous agricultural machines are: (1) Informal Group Analysis (e.g., Brainstorming); (2) Hazard Analysis and Risk Assessment (HARA); and (3) Failure Mode and Effects Analysis (FMEA).
  2. Replicability is the main advantage of FMEA and HARA, while cost effectiveness is the main advantage of Informal Group Analysis.
  3. Subjectivity and the requirement for prior knowledge (data) are the main weaknesses of FMEA, HARA, and Informal Group Analysis when applied to novel and revolutionary autonomous agricultural machines.
  4. Industry professionals do not perceive current risk assessment and hazard analysis techniques as robust and effective methods capable of ensuring the safety of autonomous agricultural machines.

Overall, autonomous agricultural machines have the potential to revolutionize modern agriculture by increasing efficiency, reducing labor and operational costs, and enhancing overall productivity. Nevertheless, these benefits come with an inherent need for rigorous safety measures. Without such robust safety systems, their widespread adoption could potentially result in significant risks. These risks encompass not only the safety of farm operators and workers but also extend to potential environmental hazards and economic setbacks. Therefore, given the importance as well as the complexity of the subject matter, collaboration between interdisciplinary teams with expertise in robotics, AI, agriculture, and safety engineering is urgently needed to comprehensively address autonomous agricultural machine safety concerns.

Acknowledgments

This research was supported with funding provided by the Central States Center for Agricultural Safety and Health (CS-CASH). Open Access was supported by the SAFER AG project supported by the intramural research program of the U.S. Department of Agriculture, National Institute of Food and Agriculture, proposal # 2022-07106 / accession # 1029426. The findings and conclusions in this publication have not been formally disseminated by the U.S. Department of Agriculture and should not be construed to represent any agency determination or policy.

References

Aby, G. R., & Issa, S. F. (2023). Safety of automated agricultural machineries: A systematic literature review. Safety, 9(1), 13. https://doi.org/10.3390/safety9010013

Aby, G. R., Issa, S. F., & Chowdhary, G. (2024). Safety risk assessment of an autonomous agricultural machine. J. Agric. Saf. Health, 30(1), 1-15. https://doi.org/10.13031/jash.15756

Dillman, D. A. (2011). Mail and Internet surveys: The tailored design method - 2007 update with new Internet, visual, and mixed-mode guide. John Wiley & Sons.

Drewry, J. L., Shutske, J. M., Trechter, D., Luck, B. D., & Pitman, L. (2019). Assessment of digital technology adoption and access barriers among crop, dairy and livestock producers in Wisconsin. Comput. Electron. Agric., 165, 104960. https://doi.org/10.1016/j.compag.2019.104960

Gerberich, S. G., Gibson, R. W., French, L. R., Lee, T.-Y., Carr, W. P., Kochevar, L.,... Shutske, J. (1998). Machinery-related injuries: Regional Rural Injury Study-I (RRIS-I). Accid. Anal. Prev., 30(6), 793-804. https://doi.org/10.1016/S0001-4575(98)00032-3

ISO. (2016). ISO 18497.2: Agricultural machinery and tractors - Safety of highly automated agricultural machines - Principles for design.

Lee, T.-Y., Gerberich, S. G., Gibson, R. W., Carr, W. P., Shutske, J., & Renier, C. M. (1996). A population-based study of tractor-related injuries: Regional Rural Injury Study-I (RRIS-I). J. Occup. Environ. Med., 38(8), 782-793. https://doi.org/10.1097/00043764-199608000-00014

Lytridis, C., Kaburlasos, V. G., Pachidis, T., Manios, M., Vrochidou, E., Kalampokas, T., & Chatzistamatis, S. (2021). An overview of cooperative robotics in agriculture. Agronomy, 11(9), 1818. https://doi.org/10.3390/agronomy11091818

Müller, M., Ghasemi, G., Jazdi, N., & Weyrich, M. (2022). Situational risk assessment design for autonomous mobile robots. Procedia CIRP, 109, 72-77. https://doi.org/10.1016/j.procir.2022.05.216

Murphy. (1999). Strategy: Engineering for hazard and injury prevention and control - Chapter 9. In Engineering for injury prevention.

Oliveira, L. F., Moreira, A. P., & Silva, M. F. (2021). Advances in agriculture robotics: A state-of-the-art review and challenges ahead. Robotics, 10(2), 52. https://doi.org/10.3390/robotics10020052

Rathour, S. S., Ishigooka, T., Otsuka, S., & Martin, R. (2020). Runtime active safety risk-assessment of highly autonomous vehicles for safe nominal behavior. No. 2020-01-0107. Proc. WCX SAE World Congress Experience. SAE International. https://doi.org/10.4271/2020-01-0107

Rial-Lovera, K. (2018). Agricultural Robots: Drivers, barriers and opportunities for adoption. Proc. 14th Int. Conf. on Precision Agriculture.

Shamshiri, R. R., Weltzien, C., Hameed, I. A., Yule, I. J., Grift, T. E., Balasundram, S. K.,... Chowdhary, G. (2018). Research and development in agricultural robotics: A perspective of digital farming. Int. J. Agric. Biol. Eng., 11(4). https://doi.org/10.25165/j.ijabe.20181104.4278

Shutske, J. M., Sandner, K. J., & Jamieson, Z. (2023). Risk assessment methods for autonomous agricultural machines: A review of current practices and future needs. Appl. Eng. Agric., 39(1), 109-120. https://doi.org/10.13031/aea.15281

Sourial, N., Wolfson, C., Zhu, B., Quail, J., Fletcher, J., Karunananthan, S.,... Bergman, H. (2010). Correspondence analysis is a useful tool to uncover the relationships among categorical variables. J. Clin. Epidemiol., 63(6), 638-646. https://doi.org/10.1016/j.jclinepi.2009.08.008

Tucan, P., Gherman, B., Major, K., Vaida, C., Major, Z., Plitea, N.,... Pisla, D. (2020). Fuzzy logic-based risk assessment of a parallel robot for elbow and wrist rehabilitation. Int. J. Environ. Res. Public. Health, 17(2), 654. https://doi.org/10.3390/ijerph17020654

Woelfel, J. l. (2022). FWEDA: Cal/OSHA Delivers blow to autonomous tractors with petition denial. Farm Equiptment, Manufacturer News. Retrieved from https://www.farm-equipment.com/articles/20472-fweda-cal-osha-delivers-blow-to-autonomous-tractors-with-petition-denial

APPENDIX

Table A1. ASABE committees that were included in survey distribution.
ASABE Committee Names
ESH-01 Ergonomics, Safety, and Health Executive / Oversight (ESH-01)
ESH-03 Ergonomics, Safety, and Health Standards Oversight
ITSC-312 Machine Vision (ITSC-312)
ITSC-318 Mechatronics & Robotics (ITSC-318)
ITSC-348 Electromagnetics & Spectroscopy (ITSC-348)
ITSC-353 Instrumentation & Controls (ITSC-353)
M-157 SMV Technologies Ergonomics, Safety and Health (M-157)
MS-01 Machinery Systems Executive (MS-01)
MS-02 Machinery Systems Steering (MS-02)
MS-03 Machinery Systems Standards Oversight (MS-03)
MS-03/2 Farm Materials Handling and Transport (MS-03/2)
MS-23 Tractors & Mach. for Ag & Forestry & US TAG ISO/TC 23
MS-23/19 Ag Electronics & US TAG ISO/TC 23/SC 19 (MS-23/19)
MS-23/19/8 Safety and security (MS-23/19/8)
MS-23/2/2 ATSC ROPS Subcommittee (MS-23/2/2)
MS-23/3 Ag Mach. - Safety and Comfort & US TAG ISO/TC 23/SC3 (MS-23/3)
MS-23/4 Tractors and US TAG ISO/TC 23/SC 4 (MS-23/4)
MS-23/4/1 Agricultural Equipment Braking (MS-23/4/1)
MS-23/4/4 Tractor & Implement Hydraulics (MS-23/4/4)
MS-23/4/5 Tractor Implement Interface/PTO (MS-23/4/5)
MS-45 Soil-Plant-Machine Dynamics (MS-45)
MS-54 Precision Agriculture (MS-54)
MS-58 Agricultural Equipment Automation (MS-58)
MS-60 Unmanned Aerial Systems (MS-60)
P-126 1/4 Scale Tractor Design Competition (P-126)
P-127 Robotics Student Design Competition (P-127)
PAFS-403/1 Milk Handling Equipment (PAFS-403/1)
PRS-293 US TAG for ISO TC 293 Feed Machinery (PRS-293)
PRS-326 Machinery for foodstuffs (PRS-326)
X497 Agricultural Machinery Management Data (X497)
MS-49 Crop Production Systems, Machinery, and Logistics