Assessment of hazard recognition performance of thermal power plant workers: A case study of a combined cycle power plant

Abstract

Hazard recognition is a vital skill every thermal power plant worker needs to possess to enhance workplace accident prevention. Facilities and materials used in thermal power plants expose the workers to significant hazards. While some prior studies have identified hazards linked to thermal power plant operations and their inherent risk rankings, knowledge of the hazard recognition performance of workers is a vital indicator of the accident risk level of workers and their training needs. Unfortunately, no empirical findings on the hazard recognition of thermal power plant workers are available. Thus, this study assesses the hazard recognition performance of thermal power plant workers at a 192 MW combined cycle power plant. An online photograph-based hazard recognition test was deployed to investigate 30 participants. Hazard categories investigated include temperature, electrical, gravity, chemical, sound, mechanical, motion, and pressure-based hazards, which are founded on Haddon’s energy release theory. Results revealed that, overall, all the workers could recognize a low proportion (21%) of typical hazards in their workplaces. Workers obtained a 29% average hazard recognition performance in identifying hazards that belong to the electrical, gravity, chemical, motion, and mechanical categories. For sound, pressure, and temperature hazard categories, workers obtained an average hazard recognition performance of 8%. This research represents the first effort to empirically assess the hazard recognition performance of workers in the context of thermal power plant operations. Findings of this research offer helpful insights to safety professionals, academic researchers, and policymakers looking to improve hazard recognition and accident prevention in thermal power plants.

Keywords:

• hazard recognition
• thermal power plant
• safety performance
• safety performance
• workplace accident

1. Introduction

Thermal power plants (TPPs) have significant hazard facilities due to the nature of hazards posed by materials and equipment used during routine plant operations (CCPS, 2010). TPP workers require excellent hazard recognition ability to ensure safety. At TPP sites, employees work with highly flammable substances, toxic chemicals, high electric voltages, high pressure-temperature steam systems, heavy-duty rotating equipment, extreme noise, and noxious fumes, which pose hazardous health risks to workers. About 76% of typical TPP hazards rank as moderate-high and high-risk hazards. Safety managers evaluate the hazard recognition ability of TPP workers to make informed decisions on the safety training content given to workers (Shrivastava & Patel, 2014). Current statistics on accidents at TPPs in literature are scanty. However, reports on closely related fields, such as electric power plants, show an alarming level of accidents. According to the Occupational Health and Safety Database of the Electric Power Research Institute (EPRI), a total of 63,193 injuries and 63 fatalities occurred in 18 electric power plants in the United States (US) from 1995 to 2013 (Volberg et al., 2017). The situation for TPPs could even be worse, considering that thermal power generation is less automated, requiring more human participation, which in turn exposes workers to more hazards during work (Diao & Ghorbani, 2018). TPP operators and maintenance personnel are more exposed to various energy forms and myriad of hazards that could jeopardize their health and usually puts them at a high risk of accidents on the job (Fleming & Fischer, 2017). Thus, the hazard recognition skills of the TPP workforce should be at best to keep accidents at a minimum.

Accident prevention is the primary goal of implementing safety management systems in the industry. Despite interventions to minimize accidents, available statistics indicate that the rate of fatal and non-fatal work-related accidents and injuries across all industrial settings is alarming. Globally, 2.3 million work-related accidents occur annually, out of which an estimated 6,000 employee deaths are recorded daily (ILOS, 2014). In Ghana, a report from the Labour Department in 2000, cited by Boadu et al. (2020), indicated that 56 out of a total of 902 accidents in Ghanaian construction industries are fatal. These accidents translate into productivity losses valued at vast amounts of money. According to Driscoll et al. (2020), work-related accidents cost the global economy an estimated US$2.8 trillion annually. Endsley (1995) and Albert and Hallowell (2012) assert that a significant proportion of workplace injuries occur due to workers’ inability to correctly predict, identify, and respond to hazardous conditions in work environments. The measure of this ability or inability is called hazard recognition performance (HRP). HRP is defined as the ratio/percentage of the number of hazards a person is able to recognize divided by the total number of hazards he/she is assessed on. For example, if a person is tested with an image which contains say 10 unique hazards and he is able to identify 7 out of the 10. Then, His HRP = 0.7 or 70%. Although there is not a standard interpretation for HRP, we consider the following guideline as ok for good interpretation: 0% to 49% = unacceptable to poor, 50% = average, and 51% to 100% = good to excellent. Insufficient hazard recognition skills account for 42% of workplace accidents (Haslam et al., 2005). Thus, hazard recognition in all workplaces has to be well evaluated and given the utmost priority to achieve high safety performance.

However, published research on hazard recognition performance have primarily focused on the construction industry (Albert et al., 2014b, 2017; Carter & Smith, 2006; Jazayeri & Dadi, 2020; Jeelani et al., 2017; Perlman et al., 2014). Other studies have also focused on the mining industry (Bahn, 2013; Barrett & Kowalski, 1995; Eiter et al., 2018). In the power generation industry, available research in literature (Adam et al., 2020; Iqbal et al., 2020; Kumar et al., 2015; Shrivastava & Patel, 2014) are only focused on the identification of hazards associated with thermal power plant operations, materials and equipment. Studies which focus on researching power plant workers’ hazard recognition performance are lacking in the published body of hazard recognition literature. To the best of the researchers’ knowledge, no such study has been done in the power generation industry, more specifically, the thermal power sector. Thus, this study being one of the first efforts to investigate workers’ hazard recognition in the sector, endeavours to fill this gap. Importantly, this study will serve as a situational alert to thermal power plant safety professionals concerning the hazard recognition performance of the workers in the sector as well as the associated accident risk implication.

The remaining part of this paper is organized as follows: Section 2 is the Literature review which discusses the background theory and present study; Section 3 outlines the research objectives and study rationale; Section 4 describes details of the experimental design and method; Section 5 presents the data analysis and results; Section 6 lays out the discussions and implications of study findings; and Section 7 is the Conclusion.

2. Literature review

2.1. Hazard recognition in other industries

Hazard recognition assessment has received much interest among safety researchers, resulting in a plethora of studies in this area (Albert et al., 2014b; Bahn, 2013; Carter & Smith, 2006; Jazayeri, 2019; Perlman et al., 2014). Findings from these studies have revealed that hazard recognition performance demonstrated by workers across various industrial domains in different locations is unsatisfactory. For instance, in the mining industry, Bahn (2013) recorded an HRP of 46%, whiles Eiter et al. (2018) had 53.5%. Similar studies in the construction industry reveal an HRP of 66% (Perlman et al., 2014) and 54% (Albert et al., 2014b). To alleviate this problem of low HRP, emphasis on safety training as a means of creating hazard awareness has been the call by researchers and industrial stakeholders (Albert et al., 2020).

Nevertheless, these conventional means of training have only brought about little impact. Zuluaga et al. (2016) concluded that the usual one-size-fits-all training is ineffective in enhancing HRP. Other studies have asserted that empirical knowledge of hazard recognition performance is required to develop and implement effective hazard recognition interventions (Albert et al., 2014a; Jeelani et al., 2019; Namian et al., 2016; Zuluaga et al., 2016). This assertion has been heeded by studies such as those by Albert et al. (2020) and Jeelani et al. (2017) in the construction industry.

2.2. Hazard recognition in the thermal power generation industry

Like other industrial settings, the thermal power plant is laden with myriad safety hazards. Knowledge of these hazards will be necessary to realize effective safety performance. As in other sectors, as previously discussed, several research efforts have been made to study these existing and potential hazards in the operations of TPPs. First, concerning the identification of dangers connected to the various plant systems, researchers such as Adam et al. (2020), Iqbal et al. (2020), and Kumar et al. (2015) have conducted studies on thermal power plants to identify the various hazards and their associated accident risk on the health and safety of workers. They reported specific threats which are pertinent to significant sections of the plant as follows:

• Water demineralization plant—Chemical burn by spillage, high noise, ammonia leakage from the storage tank or pipeline.

• Boiler system—Explosion, burns from contact with hot water, steam, or hot surface, slip, trip and fall from various height levels during routine operational work, maintenance or inspection.

• Generator and turbine—Explosion in a turbine due to cooling system failure, fire on cooling oil, high noise levels from a running turbine, pumps, etc., exposure to high temperature and pressure steam, exposure to hazardous mineral oils (control or hydraulic).

• Switchyard—Electric shock and electric burns due to unprotected high voltage equipment, maintenance or inspection of electrical panels in switch yard, fire on transformer;

• Fuel Storage—Fire hazard on fuel storage tank due to fuel leak, etc.

All hazards do not have equal gravity as each poses a different weight of risk both to workers’ safety and the power generation facility. Hazards are ranked purposely for prioritization concerning their management (Purohit et al., 2018). High-risk threats require immediate attention, more resource allocation, and emphasis on communication. Given this, Shrivastava and Patel (2014) conducted a study on two thermal plants in India to identify and classify associated hazards. Their reports revealed the distribution of risk levels posed by the 33 significant hazards identified by their study as: Low − 3%, Moderate − 21%, Moderate to High − 40%, and High risk − 36%. These results corroborate the classification of the industry as a significant hazard industry (CCPS, 2010; FLPFI, 2018).

So far, hazard-related research in the power generation industry focuses on identifying typical dangers associated with thermal power plant operations (Adam et al., 2020; Iqbal et al., 2020; Kumar et al., 2015; Shrivastava and Patel, 2014). It is highly commendable that consistent studies have been carried out into the investigation of thermal power plant operations and safety practices to identify the types and sources of hazards in the sector. Consequently, awareness of their existence will help stakeholders manage such threats. Unfortunately, an effort has yet to investigate and explore the hazard recognition performance of power plant workers. Implications are that, though the management of power plants that are aware of these hazards have implemented control interventions necessary for the safety of their workers, if the workers directly exposed to these hazards cannot recognize them, then, it makes them liable to injury. Therefore, assessing the hazard recognition performance of TPP workers is well needed.

This study empirically assesses TPP workers’ ability to recognize existing and potential hazards in their work environment and processes. Twin City Energy 192 MW Combined Cycle Power Plant (CCPP), located in the Aboadze Power Enclave, Ghana, is the case study for this research. HRP of the plant workers is assessed using an online photograph-based hazard recognition test. Computations of category-wise HRP is analyzed to determine hazard categories that workers are more proficient in recognizing.

3. Research objectives and study rational

Research has revealed that inadequate hazard recognition performance is the reason for industry accident occurrences. There is insufficient hazard recognition in the thermal power generation industry based on the injury and fatality statistics available (Kumar et al., 2015; Ünsar & Süt, 2015; Volberg et al., 2017). With this assertion, there is the need to empirically prove inadequate hazard recognition performance in the thermal power generation industry. More importantly, knowledge and understanding of workers’ hazard recognition proficiency patterns would be essential for developing and implementing effective training interventions. This study focuses on an exploratory investigation of the hazard recognition performance of TPP workers across various hazard categories—based on Haddon’s energy-release theory. This theory explains how well workers can recognize hazards and which categories they are more proficient in identifying and those they are less proficient in recognizing. In the first place, an effort like this will furnish the thermal power generation industry with the knowledge of the hazard recognition proficiency of its workers. Also, such a study will provide profound empirical insights into the bottleneck areas in the safety performance of the power generation industry, aiding in the prioritization of intervention efforts. For instance, if workers are generally less proficient in recognizing electrical hazards, emphasis will be placed on enhancing them. The study can provide the industry with evidence of the latent accident and injury risk levels exposed to workers due to their hazard recognition performance level. Hence, appropriate interventions could be developed and implemented as a proactive approach for higher worker and operational safety assurance.

4. Research method

Jazayeri and Dadi (2020) selected six (6) images in their online hazard recognition test questionnaire used to test journeymen and apprentice electricians in the US. Jamil Uddin et al. (2020) and Albert et al. (2020) used sixteen (16) images in their respective construction hazard recognition studies but tested each participant with only two (2) images randomly picked from the 16 images in their in-person test.

In reference to these studies, a hazard recognition activity was first designed and deployed online using ten (10) images from a 192 MW CCPP, which was used as the case study in order to accomplish the research objectives. Given the voluntary nature of the study, 10 was decided as an optimum number of images to cover all the hazard categories and maximize study participation (Jazayeri & Dadi, 2020). Though there are numerous TPP globally, choosing a typical case study will provide results that can be inferred to make statements on the hazard recognition performance of other TPP industries. The case hazards were classified based on Haddon’s energy release theory. This theory is on account of the underlying source of energy of the hazard (Haddon, 1995). Other researchers have successfully applied this theory in the study of the construction industry (Albert et al., 2014, Jamil Uddin et al., 2020). Table 1 indicates energy sources and their definition with corresponding hazard examples on 10 hazard categories based on Haddon’s energy release theory. This study identified 8 out of the 10 hazard categories and separates the case hazards into these 8 categories, namely, gravity, motion, mechanical, electrical, pressure, temperature, chemical, and sound. Data on workers’ hazard recognition ability was collected for each category through the online photograph-based hazard recognition test involving 30 O&M staff. The overall hazard recognition performance levels of the workers were calculated and used to determine which hazard categories workers are more and/or less proficient in recognizing. The subsequent sub-sections explain in detail, the subject population, hazard recognition test design and hazard recognition measurement.

Table 1. Hazard categories based on implicit energy sources and typical examples

Hazard categories/energy sources	Definitions	Hazard examples
Gravity	Force due to the attraction of all masses to the earth’s center.	Obstacles on floor causing trip potential; Slippery surface; Humans working at height without fall protection; etc.
Motion	The change in an object or substance’s position.	Movement of mobile equipment: crane, forklift; Human body positioning: lifting, bending, twisting; etc.
Mechanical	The energy contained or transmitted by constituents of a mechanical system or a single piece of machinery.	Compressed springs; Unguarded rotating shafts, gears, pulleys; Vibrating parts of an equipment/structure; etc.
Electrical	The presence and flow of electrical charges.	Overhead power cables; Energized equipment; Broken cable insulation; Uncovered battery terminals; Unearthed equipment; etc.
Pressure	Energy resulting from fluid(liquid/gas) which has been compressed or is under vacuum.	Compressed gas cylinders; piping, hoses; Pneumatic & hydraulic equipment/systems; etc
Temperature	The measurement of thermal energy variations in objects or the environment which is basically sensed by the human body as hot or cold.	Open flame; Hot or cold surfaces; Hot or cold liquids or gases: steam, water, air; Hot or cold weather, etc.
Chemical	The energy contained in chemical substances which inherently or via a reaction can cause physical or health harm to humans, and damage to equipment or the environment.	Low oxygen levels and high carbon monoxide in confined spaces; Volatile organic compounds (VOCs): benzene, methanol; Dust; Flammable liquid, vapours, and gases; etc
Biological	Living organism that can cause physical harm like biting, stinging or diseases.	Animals; Microorganisms: viruses, bacteria, fungi; Contaminated food or water; etc.
Radiation	The energy emitted or transmitted from a radioactive source through space or a medium.	X-rays; Welding arc; Microwaves; Solar rays; etc.
Sound	The energy released when an object is vibrated by a force which is heard by the human ear as a sound	Noise from running engine, elevated pressure relief, impact noise; etc.

(Modified from Albert et al., 2014a)

4.1. Subject population

The target population was operations and maintenance (O&M) personnel. Study subjects were recruited from the front-line O&M staff of the plant. These workers are involved in various roles: field operator, control room operator, shift charge engineer, plant and environmental chemist, instrumentation and control engineer, mechanical technician/engineer, and electrical technician/engineer.

Since the case study is a small population, the census sampling approach was applied. Thus, the entire population was used as the sample (Singh & Masuku, 2014).

Participation was entirely voluntary after the provision of informed consent by the participants. In all, 31 responses were received; however, 1 respondent only completed the demographics part of the forms but did not attempt the test itself, hence was excluded from the analysis. The excluded respondent was the only female. All participants were males with an average age of 32 years. This male-biased demographic typifies the traditional male dominance and female underrepresentation of the technical workforce of TPPs. This is due to the usually harsh, high risk and physically demanding work environment. Also, few women have the right technical education (IEA, 2022; Lehmann et al., 2021; Reitenbach, 2015). According to the US online job listing firm Zippia, based on a database of 30 million profiles, 5.1% of all power plant operators are women, while 94.9% are men (Zippia, 2022). Out of the 30 participants used, 11 were from the maintenance department, while 19 were from the operations department. All the subjects had attained tertiary-level education, whereby 86.7% had their formal educational training in engineering and 13.3% in science. Their years of experience working in the thermal power generation industry ranged from 1 to 10 years. All participants responded that they had received safety induction and other training in hazard recognition. Table 2 presents demographics of the participants.

Table 2. Demographics of participants

Number of participants from operations	Number of participants from maintenance	Average age (years)	Average working experience (years)	Number of male participants	Number of female participants
19	11	32	4.1	30	0

4.2. Hazard recognition test design and deployment

The selection of images to be used in the test began with a total stock of about 1000 photographs. Majority were pictures taken from real work operations in power plants. A few pictures were simulated work activities. Photographs were selected based on the significant hazards identified in literature; ten (10) photos were selected in order to ensure all the main work/operational sections or systems with their main associated hazards are represented. Although it is a common practical situation that the presence of one hazard further introduces other hazards, these consequential hazards are either subtle or obvious upon observation. Obvious consequential hazards are most likely to affect the recognition of the main hazard of interest. Thus, the hazards situation images were carefully selected to avoid cross effect. Only images which depicted clear and distinct hazards were selected. The final 10 photos were further processed in Adobe Photoshop software to add specific key hazards which were either underrepresented or unavailable in the images. For example, in Figure 2, smoke and fog effects were used to create high-pressure steam hazards around the steam turbine, and a section of lagged steam line was replaced with uninsulated pipe to create a hot surface hazard (temperature hazard category) as indicated by arrow 2 and arrow 1, respectively. Also, the steel barricade (inside the yellow oval in Figure 1) at the edge of the grating platform was removed to create fall-from-height hazards (gravity hazard category). In all, a total of 48 individual potentially identifiable hazards falling under 8 hazard categories were represented in the 10 photographs. The categories are: gravity, motion, mechanical, electrical, pressure, temperature, chemical, and sound. All the hazard categories apart from sound and temperature are easily recognized visually. However, according to Albert et al. (2020), visual cues promote systematic hazard recognition. Therefore, emphasis was placed on incorporating visual cues which visually insinuate sound and temperature hazards in pictures which contain sound and temperature hazard categories in order to reduce this bias. For instance, elaborated thick clouds of smoke from and around the exhaust of a diesel engine were used to depict a running engine, hence the presence of loud noise and heat or hot surfaces. A master list of these 48 potentially identifiable unique hazards was made against which responses from participants were evaluated. The test survey was then designed in Google forms. Link to the forms was sent to the participants via WhatsApp or e-mail.

Figure 1. Steam turbine and generator scene photograph before editing.

Figure 2. Sample edited photo used for hazard recognition measurement.

4.3. Hazard recognition measurement

To empirically measure the hazard recognition performance of study participants, a protocol and the relevant metric, was developed. A commonly used measurement method for hazard recognition performance is the hazard recognition index (HR _index), which is the ratio of the number of hazards identified by a person to the total number of hazards present in a particular situation (Carter & Smith, 2006; Jazayeri, 2019) or this ratio expressed in a percentage (Albert et al., 2020; Jamil Uddin et al., 2020; Jeelani et al., 2019).

For this study, the HR_index of each participating worker was calculated as the percentage proportion of identified hazards using Equation 1(1) $\mathit{H}{R}_{\mathit{index}}=\frac{H_i}{H_t}\times 100$(1) as already mentioned. For example, if a person is presented with a scene depicting a total of five hazards and upon his examination is able to identify three of them, then his HR_index = 3/5 × 100 = 60%.

(1) $\mathit{H}{R}_{\mathit{index}}=\frac{H_i}{H_t}\times 100$(1)

Where:

H_i = the number of hazards identified by worker.

H_t = Total identifiable hazards, i.e., based on predetermined hazards identified by researchers & additional hazards identified by participants through the study.

A total of 10 images were chosen for the test. All these images captured hazards that cover all the different categories previously identified. Each participant was presented with all the images and required to identify the hazards in each image.

Based on equation 1(1) $\mathit{H}{R}_{\mathit{index}}=\frac{H_i}{H_t}\times 100$(1) , the hazard recognition performance of each participant with respect to a particular hazard category ($\mathit{H}{R}_{\mathit{ij}}$), is measured by applying equation 2(2) $\mathit{H}{R}_{\mathit{ij}}=\frac{{\mathit{H}}_{\mathit{ij}}}{H_j}\times 100$(2) .

(2) $\mathit{H}{R}_{\mathit{ij}}=\frac{{\mathit{H}}_{\mathit{ij}}}{H_j}\times 100$(2)

Where:

$H_{\mathit{ij}}$ = number of hazards in hazard category j correctly identified by participant i in all images examined.

$H_j$ = total number of potentially identifiable hazards in hazard category j present in all images examined.

After obtaining the $\mathit{H}{R}_{\mathit{ij}}$ of each participant, the overall hazard recognition performance of each participant ($\mathit{H}{R}_{\mathit{iT}}$) across all hazards, was computed as the average performance across all hazard categories in all images examined using equation 3(3) $\mathit{H}{R}_{\mathit{iT}}=\frac{{\sum }_{j=1}^n\mathit{H}{R}_{\mathit{ij}}}{\mathit{m}}\mathit{\hspace{0.17em}},\mathrm{j}=\mathit{\hspace{0.17em}}1,\mathit{\hspace{0.17em}}2,\mathit{\hspace{0.17em}}\dots \dots \mathrm{n}$(3) .

(3) $\mathit{H}{R}_{\mathit{iT}}=\frac{{\sum }_{j=1}^n\mathit{H}{R}_{\mathit{ij}}}{\mathit{m}}\mathit{\hspace{0.17em}},\mathrm{j}=\mathit{\hspace{0.17em}}1,\mathit{\hspace{0.17em}}2,\mathit{\hspace{0.17em}}\dots \dots \mathrm{n}$(3)

Where:

$\mathit{H}{R}_{\mathit{iT}}$ = Overall hazard recognition performance of each participant across all hazards.

$\mathit{H}{R}_{\mathit{ij}}$= Hazard recognition performance of participant i in hazard category j in all images examined.

m = total number of hazard categories across which participant has been tested (i.e., eight for this study).

n = hazard category in which participant has been tested (e.g., temperature, gravity, etc.).

HR _index could vary from 0% when no hazards are identified and up to 100% when all hazards are identified. Carter and Smith (2006) asserted that although H_t cannot possibly account for all identifiable hazards completely, this methodology furnishes a kind of benchmark which workers can be purposefully compared to.

4.4. Data pre-processing

Responses of individual participants were extracted as follows: For each photo, the responses were compared to the master hazard list. The number of correctly identified hazards by each participant were manually tallied under the predefined hazard category. The total count is equal to the number of hazards recognized in that category. Using Equation 2(2) $\mathit{H}{R}_{\mathit{ij}}=\frac{{\mathit{H}}_{\mathit{ij}}}{H_j}\times 100$(2) , the $\mathit{H}{R}_{\mathit{index}}$ i.e., the proportion of hazards correctly recognized and recorded for that hazard category by that particular participant was first computed. This served as the primary data from which subsequent computations and analysis were made. The overall $\mathit{H}{R}_{\mathit{index}}$ and $\mathit{H}{R}_{\mathit{index}}$ per hazard category were evaluated using equations 2(2) $\mathit{H}{R}_{\mathit{ij}}=\frac{{\mathit{H}}_{\mathit{ij}}}{H_j}\times 100$(2) and 3. The results indicate the performance levels demonstrated by each participant with respect to each of the aforementioned hazard category. The MS Excel spreadsheet was used to compute and organize the primary data (HR indices) for onward analysis, while IBM SPSS Statistics 26 software package was used for the descriptive and inferential statistical analysis of the data.

5. Data analysis and results

From the calculated HRP in each of the 8 categories per participant, descriptive statistics was computed to obtain the demonstrated performance levels for each hazard category. Table 3 summarizes the results of hazard recognition indices scored by all participants in this study with the descriptive statistics: mean, standard deviation, minimum and maximum. Also, assuming 50% as a benchmark for average performance, $\mathit{H}{\mathit{R}}_{\mathit{ij}}$ scores greater than 50% for each hazard category is included in Table 3.

Table 3. Descriptive statistics of HR performance of participants by hazard category

Hazard Category	Mean $\mathit{H}{\mathit{R}}_{\mathit{ij}}$ (%)	Std. Dev. $\mathit{H}{\mathit{R}}_{\mathit{ij}}$ (%)	Minimum $\mathit{H}{\mathit{R}}_{\mathit{ij}}$ (%)	Maximum $\mathit{H}{\mathit{R}}_{\mathit{ij}}$ (%)	% of $\mathit{H}{\mathit{R}}_{\mathit{ij}}$ > 50% (%)
Temperature	14.45	9.52	0.00	33.33	0
Electrical	29.57	22.13	0.00	75.00	17
Gravity	32.22	14.45	6.67	80.00	10
Chemical	30.00	18.89	0.00	71.43	13
Sound	1.67	9.13	0.00	50.00	0
Mechanical	24.00	22.53	0.00	60.00	20
Motion	28.89	19.05	0.00	66.67	10
Pressure	8.33	18.95	0.00	50.00	0

Notes: Participant’s Hazard Recognition index by category = $\mathit{H}{R}_{\mathit{ij}}$, Percentage of $\mathit{H}{\mathit{R}}_{\mathit{ij}}$ greater than 50% = % of $\mathit{H}{R}_{\mathit{ij}}$ > 50%.

The computational outcome of the overall hazard recognition performance $\mathit{H}{R}_{\mathit{ij}}$ across all hazard categories for each participant and the corresponding descriptive statistics summary are presented in Figure 3 and Table 4 respectively. R_n in the x-axis of Figure 3 stands for the nth respondent.

Figure 3. Bar graph indicating workers overall hazard recognition performance.

Table 4. Summary descriptive statistics of overall hazard recognition performance of all participants

Number of Subject (N)	Mean$\mathit{H}{\mathit{R}}_{\mathit{iT}}$ (%)	Median$\mathit{H}{\mathit{R}}_{\mathit{iT}}$ (%)	Std. Dev. of $\mathit{H}{\mathit{R}}_{\mathit{iT}}$ (%)	Maximum$\mathit{H}{\mathit{R}}_{\mathit{iT}}$ (%)	Minimum$\mathit{H}{\mathit{R}}_{\mathit{iT}}$ (%)
30	21.24	21.69	7.81	38.18	7.15

By inspection, the descriptive statistics in Table 3 suggest differences in HRP for the various hazard categories. Nevertheless, these supposed differences could just be random and not statistically significant. However, in order to make statistically right inferences from a sample statistic, knowledge of the sampling distribution is required so that the appropriate inferential test methods can be adopted and appropriately applied (Mooney & Duval, 1993). Therefore, the data in each category was first assessed for normality and homogeneity of variance. Since the sample size is less than 50, Kolmogorov–Smirnov (with Lilliefors correction) and Shapiro–Wilk normality tests were used for the normality tests, while Levene’s test was used to test for homogeneity of variance among the different hazard categories based on different measures of central tendency at p = 0.05 significance (Table 6) (Ghasemi & Zahediasl, 2012).

Table 5. Kolmogorov–Smirnov and Shapiro–Wilk normality test results

Hazard category	Kolmogorov–Smirnov^a			Shapiro–Wilk
	Statistic	df	Sig.	Statistic	df	Sig.
Chemical	0.197	30	0.004	0.912	30	0.017
Electrical	0.248	30	0.000	0.861	30	0.001
Gravity	0.162	30	0.044	0.918	30	0.024
Mechanical	0.237	30	0.000	0.830	30	0.000
Motion	0.359	30	0.000	0.740	30	0.000
Pressure	0.503	30	0.000	0.452	30	0.000
Sound	0.539	30	0.000	0.180	30	0.000
Temperature	0.359	30	0.000	0.740	30	0.000

Notes: a. Lilliefors Significance Correction.

Table 6. Levene’s homogeneity of variances test results

Levene’s Test of Equality of Error Variances ^a,b
Type of Levene Statistic (F)	F	df1	df2	Significance
F Based on Mean	6.984	7	232	0.000
F Based on Median	4.866	7	232	0.000
F Based on Median and with adjusted df	4.866	7	180.602	0.000
F Based on trimmed mean	7.210	7	232	0.000

Notes: df₁ = group degree of freedom; df₂ = total degree of freedom.

The perceived difference in HRP was tested based on the following hypothesis:

H₀:

The hazard recognition performance (HRP) of TPP workers across all eight hazard categories are the same.

H_A: The hazard recognition performance (HRP) of TPP workers across all eight hazard categories are the different.

To assess the statistical significance of the observed differences in HRP across the various hazard categories, both one-way analysis of variance (ANOVA) and Kruskal–Wallis H test techniques could be used based on the null hypothesis that the HRP of TPP workers across all eight hazard categories are the same (Montgomery, 2020). ANOVA is usually used to test for significant differences in the averages of three or more groups. ANOVA can only be applicable if two main conditions about the groups are met, (1) groups are normally distributed, and (2) the variances among the groups to be compared are approximately the same/homogeneous (Montgomery, 2020; Pallant, 2013). Kruskal–Wallis H test on the other hand, is a non-parametric alternative to ANOVA. It is a rank based “distributionless” test which compares the medians of more than two groups of continuous or ordinal variables (Field, 2017; Pallant, 2013). Thus, unlike ANOVA, Kruskal–Wallis H test neither requires sample distribution to be normal nor homogeneity of variances of the groups being compared. For both ANOVA and Kruskal Wallis H tests, if the p-value (significance) is less than the chosen alpha value (α), which is usually 0.05 (95% confidence interval), then the researcher obtains the significant statistical proof to reject the null hypothesis (H₀) and accept the alternate hypothesis (H_A), and vice versa. Essentially, both tests can only tell you that at least two groups are different from each other. Unfortunately, they cannot indicate which groups are different. Hence, post-hoc analysis is needed in order to identify which groups are different (Field, 2017; Montgomery, 2020).

The results of both normality and homogeneity of variance tests presented in Error! Not a valid bookmark self-reference. and Table 6 respectively below, clearly indicated that all the data were not normally distributed and the variances were heterogeneous (i.e., all significance, p-values <0.05), hence the use Kruskal–Wallis test.

Tests the null hypothesis that the error variance of the dependent variable is equal across groups. a,b

1. Dependent variable: HR Performance

2. Design: Intercept + Hazard Category

From Tables 5 and 6, it could be seen that the distribution of the data is not normal, and the between group variances are heterogenous/significantly different, hence conditions for ANOVA were not met (Montgomery, 2020). With the conditions for ANOVA violated, the Kruskal–Wallis H test—the nonparametric one-way ANOVA—was therefore adopted to test for the universal null hypothesis of no difference among the groups (Gibbons & Chakraborti, 2003; Pallant, 2013; Ruxton & Beauchamp, 2008). Table 7 presents the summary results of the Kruskal–Wallis H test by which the hazard recognition performance data was subjected to.

Table 7. Kruskal–Wallis test summary of hazard recognition performance by category

Independent-samples Kruskal–Wallis test summary
N	240
H	82.618^a
Degree Of Freedom	7
Asymptotic Sig. (2-sided test), p-value	0.000

Notes: a. The test statistic is adjusted for ties; N = total number of observations; H = test statistic value.

The null hypothesis was therefore rejected based on a significance value of 0.000 from the Kruskal–Wallis test, as found in Table 7. Hence, there exist a statistically significant differences among the HR performances with respect to the eight hazard categories. Nevertheless, where the exact differences exist remains unknown. Thus, a multiple pairwise post-hoc analysis was required in order to understand which pairs of hazard categories were significantly different from each other. Therefore, the Games–Howell post-hoc multiple comparison test was adopted because it does not assume normality of data and equality of variance among groups (Games & Howell, 1976). However, it produces narrow confidence limits and keeps the experimental-wise error rate (EER) under control (Shingala & Rajyaguru, 2015)2015, thus maintaining high statistical power (Sauder & DeMars, 2019). Furthermore, bootstrapping with 1000 resamples were adopted, given the robustness of this approach against data with distribution irregularity and normality violation (Mooney & Duval, 1993). From the Bias-Corrected accelerated bootstrap confidence interval for each pairwise comparison, the difference in means was determined to be significant if the confidence interval does not include zero or both confidence limits do not intersect at zero (0) (Jamil Uddin et al., 2020; Mooney & Duval, 1993). Results of the Games–Howell pairwise post-hoc test of 28 unique comparisons performed via bootstrapping are presented in Table 8

Table 8. Pairwise comparisons of hazard recognition performance levels across the eight hazard categories

S/N	Pairwise comparison	Mean difference $\left({\bar{\mathit{\mu }}}_{\mathit{i}}-{\bar{\mathit{\mu }}}_{\mathit{j}}\right)$	Standard error	Confidence interval
				LL	UL
1	Temperature-Electrical **	−15.12	4.529	−25.234	−6.665
2	Temperature-Gravity **	−17.78	3.200	−23.775	−11.985
3	Temperature-Chemical **	−15.56	3.866	−23.679	−7.799
4	Temperature-Sound **	12.78	2.464	7.538	17.206
5	Temperature-Mechanical **	−9.56	4.596	−18.541	−1.120
6	Temperature-Motion **	−14.44	3.702	−21.872	−7.126
7	Temperature-Pressure	6.11	3.792	−1.723	13.194
8	Electrical-Gravity	−2.66	4.930	−11.535	7.156
9	Electrical-Chemical	−0.44	5.323	−11.517	10.550
10	Electrical-Sound **	27.90	4.417	19.655	36.995
11	Electrical-Mechanical	5.57	5.867	−5.391	16.571
12	Electrical-Motion	0.68	5.231	−9.324	12.249
13	Electrical-Pressure **	21.23	5.268	10.992	32.701
14	Gravity-Chemical	2.22	4.415	−7.125	11.988
15	Gravity-Sound **	30.56	3.184	24.201	36.692
16	Gravity-Mechanical	8.22	4.970	−1.357	16.819
17	Gravity-Motion	3.34	4.235	−5.064	11.963
18	Gravity-Pressure **	23.89	4.231	16.073	32.296
19	Chemical-Sound **	28.34	3.800	21.095	35.320
20	Chemical-Mechanical	6.00	5.356	−4.232	15.711
21	Chemical-Motion	1.11	4.733	−8.218	10.410
22	Chemical-Pressure **	21.67	4.739	12.758	29.910
23	Sound-Mechanical **	−22.33	4.537	−30.443	−14.072
24	Sound-Motion **	−27.22	3.678	−34.564	−19.877
25	Sound-Pressure	−6.67	3.809	−14.471	1.190
26	Mechanical-Motion	−4.89	5.353	−15.593	6.159
27	Mechanical-Pressure **	15.67	5.455	3.532	27.272
28	Motion-Pressure **	20.56	4.774	10.375	29.999

Notes: ** indicates pairwise comparison revealed a statistically significant difference in hazard recognition performance levels; LL = lower limit; UL = upper limit.

Finally, towards establishing hazard categories (HCs) which workers were more proficient in recognizing and those they were less proficient in recognizing, further clarity and insight were drawn from the results in Table 8. Thus, HCs have been segregated into Group 1 and Group 2 in Table 9 based on the statistical similarity of their mean values—no significant difference in respective means when group members are compared. Also, for each group, the mean HRP scores for each hazard category and the group mean HRP has been presented as well in Table 9.

Table 9. Summary of pairwise comparisons

Group 1		Group 2
Hazard category	Mean $\mathit{H}{\mathit{R}}_{\mathit{ij}}$ (%)	Hazard category	Mean $\mathit{H}{\mathit{R}}_{\mathit{ij}}$ (%)
Electrical	29.58	Sound	1.67
Gravity	32.22	Pressure	8.33
Chemical	30.00	Temperature	14.44
Motion	28.89
Mechanical	24.00
Mean Group HRP	28.94	Mean Group HRP	8.12

Notes: $\mathit{H}{\mathit{R}}_{\mathit{ij}}$ = mean hazard recognition performance by category, HRP = hazard recognition performance.

6. Discussions and implications of study findings

Findings from the study furnish valuable insights that safety managers can use to improve safety performance in the thermal power generation industry via hazard recognition improvement. The mean overall hazard recognition performance of 21.14% demonstrated by the participants in this study corroborates the outcomes of previous research in various industry sectors that workers generally are unable to recognize a good number of safety hazards. Furthermore, compared to overall HRPs reported in previous studies such as 46% (Bahn, 2013), 53.5% (Eiter et al., 2018) in mining, 66% (Perlman et al., 2014), 54% (Albert et al., 2014) in construction, 21.14% is very low and should be a major concern for the thermal power generation sector. Past researchers have linked frequent accidents and injuries to low HRP of workers (Jazayeri, 2019; Purohit et al., 2018).

The inability of the TPP workers to recognize 79% of the typical hazards implies that they are at a very high risk of involving in accidents leading to injuries or even fatality. For instance, Kumar et al. (2015) found that overconfidence, which is a result of poor hazard recognition on the side of experienced workers at an Indian power plant, was the leading cause of accidents at the plant, resulting in a total of 59 worker injuries. Haas et al. (2017) and Perlman et al. (2014) also made a similar assertion in their study. Furthermore, Ünsar and Süt (2015) reported 941 accidents and 15 deaths involving Turkish power generation plant workers over a period of 8 years. Also, from the repository of the Occupational Health and Safety Database of the Electric Power Research Institute (EPRI), Volberg et al. (2017) reported that 63,193 accidents and 21 deaths happened across 18 United States power plants spanning 18 years (1995–2013). Therefore, appropriate intervention should be deployed to avert this potentially catastrophic situation.

Training is the primary intervention used by employers to improve the hazard recognition performance of workers (Jeelani et al., 2017). However, it has been empirically proven that the usual one-size-fits-all training administered through classroom lectures, videos, and books is ineffective (Albert et al., 2017; Namian et al., 2016). This is because these traditional methods of training are passive, less engaging and general. More focused engagement training has been found to produce a positive influence on workers’ hazard recognition performance (Zuluaga et al., 2016). Also, different trainee groups require different types of training content and methods as factors like level of experience and current job affects the influence of training on hazard recognition. Zhu et al. (2022) reported that their telegram-based chatbot training significantly impacted the hazard recognition of novice onsite construction workers and off-site construction professionals more than highly experienced onsite practitioners. Therefore, they emphasised that training content should be matched with trainee.

Thus, personalize training, an example of strategic safety performance improvement intervention based on knowledge and understanding of workers’ hazard recognition, is more effective (Albert et al., 2014a; Jeelani et al., 2017; Namian et al., 2016; Zuluaga et al., 2016). Though all the participants have received both general safety induction and hazard recognition training, the poor HRP demonstrated by the workers suggests the ineffectiveness of the training they have received. Therefore, insights from this study will provide the essential inputs for developing and implementing an effective training intervention to improve the HRP of workers. Furthermore, this study has revealed that workers are more proficient in recognizing certain hazard categories than others. Specifically, workers exhibited relatively higher proficiency in recognizing the electrical, gravity, chemical, motion, and mechanical hazard categories with an average HRP of 29%. On the other hand, they were less proficient in recognizing sound, pressure, and temperature hazard categories, with an average HRP index of 8%.

Moreover, high noise and extreme pressure hazards are characteristic of TPP operations. Nevertheless, the majority failed to recognize hazards in these categories. For instance, only one noise hazard situation (Mean = 1.67%) was recognized by one participant out of the 30. This confirms the assertion that the familiarity of a worker with certain hazards and work environments causes less sensitivity to those hazards (Haas et al., 2017; Perlman et al., 2014), resulting in failure to recognize them. Since burns resulting from temperature hazards are responsible for most severe injuries and even fatalities in TPP operations, special attention ought to be given to hazard categories that workers are less proficient in recognizing, especially the most familiar ones. That said, it is worth noting that even in the categories where workers showed more proficiency (i.e., 29%), the performance level is still unsatisfactory. These findings reveal workers’ inadequate hazard recognition performance and suggest that the thermal power plant sector pays attention to developing and implementing new interventions focusing on all hazard categories.

7. Conclusion

This study seeks to empirically assess the ability of thermal power plant workers in recognizing existing and potential hazards in their work environment and processes. Findings revealed that TPP workers, on average, were able to recognize 21% of hazards across all 8 hazard categories. This suggests that TPP workers’ ability to recognize workplace hazards is far from acceptable. Consequently, TPP workers are at a high risk of injury or fatality from exposure to or interaction with hazards on their job. Furthermore, TPP workers exhibited relatively high proficiency in recognizing electrical, gravity, chemical, motion, and mechanical hazard categories at an average performance of 29%. However, workers demonstrated relatively less proficiency in recognizing sound, pressure, and temperature hazard categories at an average performance of 8%.

The study provides essential insights based on which TPP workers’ hazard recognition can be enhanced. Emphasis and attention may be given to the improvement of hazard recognition in the hazard categories that workers are less proficient in recognizing by providing training that focuses on these hazard categories. The results revealed low performance in hazard categories where workers showed relatively higher proficiency. Hence, current efforts directed at these hazards should be maintained and increased to ensure higher performance.

While the thermal power plant industry must continue its emphasis on hazard recognition improvement programmes, attention must also be devoted to the hazard categories that workers are less proficient in recognizing, as low HRP performance can bring about unfavourable incidents and financial losses. Policymakers, researchers, and industry leaders looking to enhance safety in the thermal power generation industry could benefit from the findings of this study.

Moreover, one of the principal limitations of the research was the use of work situation photographs which may not adequately represent certain hazards in the sound and pressure categories pertinent to thermal power plant operations. Although the methods adopted furnished safe and regularized means of performance assessment and comparison, future studies may replicate the study effort but concentrate on workers observing live plant operational activities, recorded video or using immersive augmented and/or virtual reality setup.

Finally, since the study was focused on a combined cycle thermal power plant, the extent to which the results can be generalized may be limited. Nevertheless, characteristics of the case study and subject population typify thermal power plant workforce irrespective of the location. Hence, the study findings provide reliable evidence of poor hazard recognition performance of thermal power plant workers in general—which is partly due to the ineffectiveness of the current hazard recognition training programmes run in the thermal power generation sector.

CRediT authorship contribution statement

Romeo Danquah: Data curation, Methodology, Formal analysis. Gidphil Mensah: Formal analysis, Validation, Writing—review & editing. Winfred Senyo Agbagah: Validation, Writing—review & editing. Francis Davis: Conceptualization, Formal analysis, Validation, Writing—review & editing.

Acknowledgments

We are grateful to the studied thermal power plant and very thankful to all the respondents who voluntarily participated in the study.

Disclosure statement

No potential conflict of interest was reported by the authors.

Data availability statement

Data associated with this study will be made available upon request.