Classification for Hyperthermal Environments Based on a Comprehensive Score Index

Abstract

Working in hyperthermal environments can lead to heat-related illnesses. Evaluating and predicting high-temperature environments can effectively reduce heat risks and hazards. However, there is still a lack of corresponding high-temperature environment assessment methods and indicators in existing research. Moreover, traditional evaluation indicators and prediction methods have shortcomings in objectivity, accuracy, and practicality. To fill these gaps, a climate chamber was constructed to simulate different environmental conditions, and human labor experiments with 98 subjects were conducted. The ambient temperatures were set to 34 °C, 36 °C, 38 °C, and 40 °C, and the relative humidity was set to 60%, 70%, 80%, and 90%, respectively. During the experiments, the subjects’ oral temperatures, heart rates, skin temperatures, and subjective perceptions were recorded. Based on the obtained parameters of the subjects, two principal components with an explained variance of 92.131% were extracted by principal component analysis, and with the determination of weightings, a comprehensive evaluation index (F) was established and the F-score was calculated. According to the F-score, 16 different hyperthermal environments were classified into three categories through hierarchical clustering analysis and discriminant analysis, with the corresponding F-score ranges of 28.14–39.76, 39.17–45.21, and 44.13–52.39. An analysis was conducted on the value of physiological and subjective indicators to test the nature of classification.

1. Introduction

Hyperthermal environments refer to working environments where the air temperature is above 32 °C and the relative humidity (RH) is above 60% [1,2]. They are common in many workplaces [3,4]. Personnel working in hyperthermal environments will experience significant physical heat strain, while also facing health and safety risks [5]. Exposure to high temperatures can lead to heatstroke, heat exhaustion, and heat rash [6]. It has effects on heat balance [7], water balance and electrolyte metabolism [8,9], the blood circulation system [10,11], and the nervous system [12] of the human body.

Numerous studies have confirmed that working in hyperthermal environments will cause significant negative effects on the human body. Donoghue [13] described some common fever diseases among miners and stated that ventilation and cooling cannot effectively control hyperthermal environments in mines. Zhu’s [14] research showed that high temperatures can cause the sympathetic nervous system to be in an excited state, reducing the activity of thermoregulatory effectors and leading to thermal discomfort in the human body. Based on an experiment with 20 participants, Jing [15] found that high humidity, together with temperature, has a negative impact on human thermal comfort. Quelhas [16] discovered significant effects of environmental conditions on human productivity. Yang [17] set eight conditions in an experiment and found that when the temperature is above 38 °C and the humidity is above 50%, women’s long-term labor will no longer be safe. Vangelova [18] studied the blood lipid status of workers in hyperthermal environments, and found that workers who were exposed to hyperthermal environments for a long time may be more susceptible to hyperlipidemia. Chen’s research [19] showed that workers in hyperthermal environments were more prone to subjective fatigue, and fatigue levels will increase with increasing heat exposure. In addition, many countries or regions have reported deaths related to hyperthermal environments. It is reported that 423 workers in agricultural and industries have died due to exposure to hyperthermal environments [20]. Strong heat stress in summer can lead to a large number of deaths related to heat diseases [21,22]. In 2009, there were very serious heat waves. Among these heat waves in Adelaide, the mortality and incidence rate ranked third [23].

Inaccurate or untimely evaluation of hyperthermal environments may result in heat hazards or life-threatening situations for personnel. Due to individual differences in human bodies, there may also be differences in subjective perception and physiological parameters [17]. In addition, the occurrence of some heat hazards is usually short-term and acute, and treatment may be delayed if there is a lack of early warning or first aid. Therefore, it is necessary to evaluate and predict environmental conditions to develop relevant standards and make management decisions.

When evaluating and predicting hyperthermal environments, relying solely on environmental parameters cannot reflect the impact of the environment on human strain. Therefore, it is necessary to determine the heat status of the subjects based on their body indicators. Du [24] proposed safe physiological limits for human bodies, which are a rectal temperature of 38.5 °C and a heart rate of 145 beats/min. At the same time, it is also pointed that the maximum tolerance limits of human bodies are 39.4 °C and 174 beats/min. Lv [25] conducted an experiment and found that the limits of the human body for oral temperature and water loss rate were 38 °C and 1%, respectively. Nag [26,27] evaluated human tolerance limits and found that the tolerance time for people in an environment with 38–38.2 °C is 80–85 min, and estimated the safe exposure times for women, which varied from 43 min (32.0 °C ET (N))to 16 min (36.5 °C ET (N)). The typical limit for the increase in deep body core temperature is 1 °C [28,29]. The typical limit for heat storage in the body is 60 watts/hour/meter for acclimatized workers (50 watts/hour/meter for unacclimatized workers) [30].

Air temperature (T_a) and relative humidity (RH) are two environmental variables that have effects on comfort level [31]. Environmental conditions will affect labor efficiency and reduce productivity for people who are in such an environment [32]. Previous indicators have some shortcomings and deficiencies. First, as mentioned earlier, only selecting environmental parameters for environmental condition classification and prediction cannot accurately reflect the relationship between the environment and human thermal responses [30]. Second, when selecting physiological parameters, traditional indicators only consider rectal temperature and heart rate, thus ignoring the important role of other indicators.

In summary, there are some main gaps in evaluating and predicting hyperthermal environments. It is necessary to develop a comprehensive index that considers the effects of environmental conditions on the human body, and the classification method for hyperthermal environments can be innovated to take into account the correlation of physiological parameters, which is one of the key aspects of this study. Meanwhile, it is also necessary to establish a hierarchical model with specific equations and continuum results. In order to fill these gaps, this study aims to conduct human thermal strain experiments under high temperature environmental conditions. Based on the obtained experimental data, a new environmental evaluation and classified index was constructed through reliable analysis methods, and cluster analysis was used to classify 16 environmental conditions. These results provide reference points for occupational safety and health in hyperthermal environments.

2. Materials and Methods

2.1. Experiment

2.1.1. Chamber

A stainless steel chamber with a size of 5 m × 4 m × 3 m was built to simulate hyperthermal environments. The climate chamber was relatively closed, and the temperatures of the inner walls were basically unchanged. The indoor temperature and relative humidity can be controlled to preset values through a programmable controller. The heating source of the chamber was stainless steel flake-shaped heaters, and the humidity source was stainless steel humidifiers. At the top of the climate chamber, there was a sensitive probe to monitor the real-time air temperature and relative humidity, and the values were sent to the control panel. The range of temperature was from −20 °C to 85 °C, and the deviation was ≤±0.5 °C. The range of humidity was from 20% to 98%, and the deviation was ≤±3.0%. The chamber was equipped with a remote alarm function and protection system to ensure the safety of the chamber and personnel.

2.1.2. Subjects

Ninety-eight male participants were chosen as subjects. All of the subjects had good health and no disease history. Those with cardiovascular, cerebrovascular, or metabolic contraindications to exercise were excluded. Most of the subjects have lived locally for more than 1 year. The ages, heights, weights, and surface areas of the subjects were 22.1 ± 2.1 years, 174.5 ± 5.5 cm, 65.13 ± 9.17 kg, and 1.87 ± 0.13 m², respectively. All the subjects were informed of heat risk and heat disease treatment prior to the experiments. They were also asked to dress similar clothes, including short pants, socks, and sports shoes, during the experiments. All the subjects provided informed consent.

2.1.3. Parameters and Instruments

The physiological parameters and their corresponding instruments are shown in

Table 1. Parameters and instruments in the experiment.

Parameter	Instrument	Model	Range	Accuracy
Heart rate	Blood pressure monitor	HEM-7112 (Omron, Dalian, China)	40—180 beats/min	±5%
Oral temperature	Electronic thermometer	MC—347 (Omron, Dalian, China)	32.0—42.0 °C	±0.1 °C
Mean skin temperature	Infrared thermometer	MC—720 (Omron, Dalian, China)	0—50.0 °C	±0.1 °C
Blood pressure	Blood pressure monitor	HEM-7112 (Omron, Dalian, China)	0—299 mmHg	±3 mmHg
Weight	Electronic scale	TCS150 (Changxie, Dongguan, China)	0—150 kg	±1 g

.

The skin temperature (T_sk) was measured at eight sites in accordance with the formula and weighting of ISO standard 9886, as follows [33,34]:

$$\begin{array}{ll}{T}_{\mathrm{sk}}& =0.07\times T_{forehead}+0.175\times T_{rightscapula}+0.175\times T_{leftupperchest}\\ & +0.07\times T_{rightarminupperlocation}+0.07\times T_{leftarminlowerlocation}+0.05\times T_{lefthand}\\ & +0.19\times T_{rightanteriorthigh}+0.2\times T_{leftcalf}\end{array}$$

(1)

The oral temperature (T_or) measured is regarded as core temperature [35], whereas body weight and the weight of the water consumed by the subjects were also measured to calculate body weight loss [36].

2.1.4. Experiment Process

Different combinations of air temperature and relative humidity were adopted during the whole experiment, as shown in

Table 2. Environmental conditions in the experiments.

Parameter	Condition
Temperature (°C)	34				36
Relative humidity (%)	60	70	80	90	60	70	80	90
Temperature (°C)	38				40
Relative humidity (%)	60	70	80	90	60	70	80	90

.

Before each experiment, all participants were required to rest for 20–40 min to stabilize their body parameters. The initial physiological parameters of the subjects were measured after they entered the chamber. The experiments were performed in a balanced random order, with each subject randomly assigned to one of the conditions. In particular, considering the consensus on the sample size in this research area, this study arranged 6–8 subjects for each experimental condition. Treadmill exercise at 6 km/h was used to simulate heavy manual labor, and the intensity index of physical work (IIPW) was 20 [37,38]. It was considered that working in a hyperthermal environment for 100 min is the best time [39]. Therefore, during each experiment, the subjects were required to run on the treadmills for 90 min, and they had a 3 min rest for every 15 min work. Furthermore, a 2-day interval between two successive experiments was required to avoid heat acclimation. T_or, T_sk, and HR were measured during break. Additionally, the subjects were asked to fill in the subjective perception checklist [40,41]. Each subject could not stop until their physiological indexes indicated that they could not continue to work anymore or they subjectively could not suffer from the labor intensity and the hyperthermal environments, and the time when the experiment stopped was recorded.

2.2. Methodology

2.2.1. Factor Analysis

Factor analysis (FA) is commonly used for statistical analysis. It aims to describe the relationship among some indexes by using few factors. In detail, FA puts a few closely related variables into the same class; each type of variable becomes a factor, reflecting most of the original information with a few factors. It has been successfully applied in many areas [42,43,44,45,46].

In general, it is impossible for common factors to cover all the information; the part that they cannot cover is called special factors. The general form of factor analysis is as follows [47]:

$$\left\{\begin{array}{c}{X}_1-{\mu }_1=a_{11}\times F_1+a_{12}\times F_2+\cdots +a_{1m}\times F_m+{\epsilon }_1\\ X_2-{\mu }_2=a_{21}\times F_1+a_{22}\times F_2+\dots +a_{2m}\times F_m+{\epsilon }_2\\ ••••••\\ X_p-{\mu }_p=a_{p1}\times F_1+a_{p2}\times F_2+\cdots +a_{pm}\times F_m+{\epsilon }_p\end{array}\right.$$

(2)

where F₁, F₂, ⋯, F_m are called common factors of an initial variable, and ε_i (i = 1, 2,⋯, p) is special factors of variables X_i.

2.2.2. Cluster Analysis (CA)

Cluster analysis (CA) is one of the leading methods of multivariate analysis [48]. It aims to classify the collection of samples into many classes consisting of similar samples [49]. Being similar to fFA, CA is also applied in many areas [50,51,52,53,54,55].

Hierarchical cluster is used in this research, and squared Euclidian distance is used for defining distance, which is obtained by the following Equation (3) [56]:

$$d_{ij}={{\displaystyle \sum _{k=1}^p\left(X_{ik}-X_{jk}\right)}}^2$$

(3)

where d_ij is squared Euclidian distance, and X_ik and X_jk are observation values of samples.

In addition, the Ward method is chosen as the method to perform hierarchical cluster. The distance between one individual and one group can be shown as follows [57]:

$$D\left(R,P+Q\right)={\displaystyle \frac{1}{NR+NP+NQ}}\left[\left(NR+NP\right)\times D\left(R,P\right)+\left(NR+NQ\right)\times D\left(R,Q\right)-NR\times D\left(P,Q\right)\right]$$

(4)

where R, P, and Q are individuals; P + Q is incorporated as one group; and N is the number of observations for each individual.

The square sum of dispersion can be calculated as follows:

$$V={{\displaystyle \sum _{k=1}^K{\displaystyle \sum _{j=1}^J\left(X_{kj}-\overline{X_j}\right)}}}^2$$

(5)

where X_kj is the observation values for variables j (j = 1, 2, ···, J) of objects k (k = 1, 2, ···, K), and $\overline{X_j}$ is the average value of observation values for variables j in a group.

2.2.3. Fisher Discriminant Analysis (FDA)

Fisher discriminant analysis (FDA) is a classical method to extract feature and reduce dimension jointly [58,59,60]. It is a discriminant method according to the principle of variance analysis, and it has been applied in many areas [61,62,63,64,65,66,67].

Establishing a discriminant function is needed before conducting FDA. The general form of a discriminant function is as follows [59]:

$$Y=b_0+b_1\times X_1+b_2\times X_2+\dots +b_J\times X_J$$

(6)

where Y is a discriminant indicator, X_j (j = 1, 2, ···, J) is discriminant variables, b_j is the coefficients of discriminant variables X_j, and b₀ is the constant value.

2.2.4. Data Processing

In this paper, the factor analysis method, hierarchical cluster analysis method, and Fisher discriminant analysis method are used. All data were analyzed using SPSS (Statistical Package for the Social Sciences), version 26.0 (SPSS Inc., Chicago, IL, USA). SPSS has a history of over 50 years and is a professional software for statistical analysis. Its algorithm and output results have been extensively tested and validated through practical research, and have good reliability [1,3,4,5,6]. Statistical significance was set at p < 0.05. Values are reported as mean ± standard deviation.

3. Results

3.1. The Comprehensive Score Index

Three physiological indexes and two psychological indexes were chosen to propose the comprehensive index. The three physiological indexes were mean skin temperature (T_sk), oral temperature (T_or), and heart rate (HR), and the two psychological indexes were thermal sensation votes (TSV) and feelings of fatigue (FOF). T_sk, T_or, and HR are easily measured by electronic instruments, and the results can be much more precise. The skin is an important organ in temperature regulation; meanwhile, T_sk has been extensively used as a physiological indicator to assess thermal comfort [68,69,70,71]. Oral temperature can be considered as core temperature; T_or and HR are often applied in thermal comfort assessment [72]. FOF and TSV are two subjective indexes for the subjects.

3.1.1. Determination of Weightings

The value of variables of each subject were used to carry out factor analysis.

Table 3. Correlation matrix of the variables.

		Tor	HR	Tsk	FOF	TSV
Correlation	Tor	1.000	0.813	0.768	0.257	0.544
	HR	0.803	1.000	0.686	0.281	0.560
	Tsk	0.768	0.669	1.000	0.278	0.582
	FOF	0.657	0.681	0.689	1.000	0.739
	TSV	0.734	0.760	0.752	0.735	1.000
Sig.	Tor	/	0.000	0.000	0.077	0.018
	HR	0.000	/	0.000	0.066	0.017
	Tsk	0.000	0.000	/	0.052	0.011
	FOF	0.037	0.066	0.031	/	0.000
	TSV	0.027	0.026	0.021	0.000	/

shows the correlation matrix of the variables. For most of the variables, the correlation coefficients are bigger than 0.5 [73], and the corresponding value of Sig. is statistically significant. The results illustrate a significant correlation between the variables and the necessity of using FA.

On the basis of a large number of experimental data, the principal component analysis was conducted and passed the Kaiser–Meyer–Olkin (KMO) test and Bartlett’s test of sphericity. The results are given in

Table 4. KMO and Bartlett’s test of sphericity.

Kaiser–Meyer–Olkin Measure of Sampling Adequacy		0.717
Bartlett’s test of sphericity	Approx. Chi-Square	102.762
	df	10
	Sig.	0.000

. The value of KMO statistics is 0.717 and is bigger than 0.5, which indicates that the total variances of principal components extracted contribute highly to the population variances [73]. Therefore, the data from the experiment are suitable to be analyzed with FA [74]. Additionally, the value of Sig. for Bartlett’s test of sphericity is smaller than 0.001, which illustrates the significant correlation between variables.

The variance and its cumulative sum explained by each common factor are given in

Table 5. Total variance explained.

Component	Initial Eigenvalues			Extraction Sums of Squared Loadings			Rotation Sums of Squared Loadings
	Total	Variance Ratio	Cumulative Percentage	Total	Variance Ratio	Cumulative Percentage	Total	Variance Ratio	Cumulative Percentage
1	2.989	69.565	69.565	2.989	69.565	69.565	2.514	50.180	50.180
2	1.289	25.565	93.231	1.289	25.565	85.131	1.763	35.061	85.131

. The cumulative variance explained by the first two common factors is 93.231%, which is bigger than 85%, so these two common factors could explain the information covered by the original variables [75].

Table 6. Rotated component matrix.

	Component
	1	2
Tor	0.921	0.145
HR	0.912	0.161
Tsk	0.876	0.192
FOF	0.118	0.920
TSV	0.256	0.896

shows that the first common factor has a larger load on the first three variables, which are exactly the physiological indexes. T_or, T_sk, and HR have significant correlation, and they can be classified as one group. Therefore, the first common factor can be named as the objective factor. Meanwhile, the second common factor has a larger load on the last two variables, which are exactly the subjective indexes. Similarly, the second common factor can be named as the subjective factor.

Therefore, the expression of the two common factors can be obtained from

Table 7. Component score coefficient matrix.

	Component
	1	2
Tor	0.400	−0.091
HR	0.393	−0.064
Tsk	0.381	−0.057
FOF	−0.138	0.690
TSV	−0.060	0.652

and is as follows:

$$F_1=0.400\times X_1+0.383\times X_2+0.361\times X_3-0.136\times X_4-0.070\times X_5$$

(7)

$$F_2=-0.091\times X_1-0.074\times X_2-0.047\times X_3+0.590\times X_4+0.542\times X_5$$

(8)

where F₁ is the objective factor, F₂ is the subjective factor, X₁ is oral temperature (T_or), X₂ is heart rate (HR), X₃ is the mean skin temperature (T_sk), X₄ is the feeling of fatigue (FOF), and X₅ is the thermal sensation vote (TSV).

For further comprehensive evaluation, it is also necessary to weight the two common factors by weighting the contribution rate of the variance to the cumulative contribution rate as a weight to calculate the comprehensive scores. The function of the comprehensive scores is as follows:

$$F={\displaystyle \frac{0.50080}{0.85131}}\times F1+{\displaystyle \frac{0.35051}{0.85131}}\times F2=0.588\times F1+0.412\times F2$$

(9)

where F is the comprehensive score index. It is composed of five variables: oral temperature (T_or), heart rate (HR), mean skin temperature (T_sk), feelings of fatigue (FOF), and thermal sensation vote (TSV). The formula is used to calculate the score of each subject to further analyze their thermal states.

3.1.2. Validation of the Comprehensive Score Index

To verify the validity of the comprehensive score index (F), the correlation analysis was conducted with the PSI. Due to the fact that the core temperature of the human body in the experiment was obtained through oral temperature testing, the PSI was modified as follows:

$$PSI=5\times {\displaystyle \frac{T_{ort}-T_{or0}}{39.5-T_{or0}}}+5\times {\displaystyle \frac{H{R}_t-H{R}_0}{180-H{R}_0}}$$

(10)

where T_ort is the measured values of oral temperature at any time t, HR_t is the measured values of heart rates at any time t, T_or0 is the measured values of oral temperature at the initial time, and HR₀ is the measured values of heart rates at the initial time.

The correlation analysis between the comprehensive score index (F) and the modified PSI is shown in Figure 1. The correlation coefficient is 0.957, which proves the effectiveness and rationality of the comprehensive score index (F). Moreover, compared with the PSI, the comprehensive score index (F) is correlated with changes in human subjective perception, further demonstrating its effectiveness.

In summary, the correlation is summarized in

Table 8. The correlation between comprehensive score index and PSI.

Variables	Comprehensive Score Index (F)	Modified PSI	Sig.
Comprehensive score index (F)	1.000	0.957	0.000
Modified PSI	0.957	1.000

, and the significance level is p < 0.001, which well verifies the validity of the comprehensive score index (F).

3.2. Classifying the Hyperthermal Environment

The Ward method is used in this research for hierarchical cluster. The score of each subject is calculated according to the comprehensive score index F. The value of it could represent the degree of badness for the different environments.

The results of the CA for the score of F of each subject are shown in Figure 2; the subjects were working under 34 °C, 36 °C, 38 °C, and 40 °C. The red, yellow, and blue columns represent the first-, second-, and third-class environments. Most of the values of the scores in the red column are lowest; they are almost lower than 40. The value of the score in the yellow column is higher than that in the red column but lower than that in the blue column, and most of them fall within the interval of 40–45 min. The value of the score for the blue column is highest, and most of them are bigger than 45 min and even 50 min.

The value of the score increases with the increase in humidity when the temperature remains the same; also, the value of the score increases with the increase in temperature when the humidity remains the same. Both temperature and humidity have significant effects on the thermal state of the subjects.

Table 9. The classification of working conditions.

Working Conditions	Number of Red Columns	Number of Yellow Columns	Number of Blue Columns	Classification
34 °C—RH60%	5	1	0	1
34 °C—RH70%	4	2	0	1
34 °C—RH80%	6	0	0	1
34 °C—RH90%	4	2	0	1
36 °C—RH60%	5	1	0	1
36 °C—RH70%	4	2	0	1
36 °C—RH80%	4	2	0	1
36 °C—RH90%	0	4	2	2
38 °C—RH60%	2	5	0	2
38 °C—RH70%	3	4	0	2
38 °C—RH80%	1	2	3	3
38 °C—RH90%	0	0	6	3
40 °C—RH60%	2	3	1	2
40 °C—RH70%	0	1	5	3
40 °C—RH80%	0	1	5	3
40 °C—RH90%	0	0	6	3

shows the result of classification for the 16 different environments by clustering analysis. The classification of each working condition depends on the score of F of each subject within. Therefore, in the first-class environment, 34 °C—RH60%, 34 °C—RH70%, 34 °C—RH80%, 34 °C—RH90%, 36 °C—RH60%, 36 °C—RH70%, and 36 °C—RH80%. In the second-class environment, 36 °C—RH90%, 38 °C—RH60%, 38 °C—RH70%, and 40 °C—RH60%. In the third-class environment, 38 °C—RH80%, 38 °C—RH90%, 40 °C—RH70%, 40 °C—RH80%, and 40 °C—RH90%.

The value of the score of F for the first-class environment ranges from 28.14 to 41.69. For the second-class environment, it ranges from 39.17 to 45.21. Additionally, for the third-class environment, it ranges from 44.13 to 52.39. However, the result is not precise due to the existence of overlapping. This part of the data is extracted to be further analyzed by Fisher discriminant analysis (FDA).

3.3. The Further Verification for the Classification of F

The initial result of the interval of F for each class of the environment is given by clustering analysis (CA). These data are extracted to be further analyzed by Fisher discriminant analysis (FDA). According to the clustering results, FDA is performed on the parts of F whose value of the score is not overlapped. A discriminant function is established and shown in

Table 10. The coefficients of classification function.

	Classification
	1	2	3
The score of the 1st common factor	4.093	4.714	5.580
The score of the 2nd common factor	−0.949	−1.648	1.113
Constant	−148.468	−203.902	−243.139

. The data that are overlapped are further verified based on the established discriminant function.

Six data were extracted to be further verified by FDA. The six data are 41.69, 39.59, 39.76, 39.58, 39.37, and 45.00. When performing FDA, these six data are ungrouped; the statistic results are given in

Table 11. Casewise statistics for the ungrouped and part of data.

Case Number		Actual Group	Highest Group					Second Highest Group			Discriminant Scores
			Predicted Group	P(D > d|G = g)		P(G = g|D = d))	Squared Mahalanobis Distance to Centroid	Group	P(G = g|D = d)	Squared Mahalanobis Distance to Centroid	Function1	Function2
				p	df
Original	1	2	2	0.801	2	0.986	0.443	1	0.014	8.933	−0.436	−1.756
	2	1	1	0.152	2	1.000	3.768	2	0.000	20.580	−4.191	0.267
	3	1	1	0.927	2	0.984	0.152	2	0.016	8.417	−2.397	0.227
	4	1	1	0.655	2	0.999	0.846	2	0.001	14.176	−2.437	1.503
	5	1	1	0.023	2	1.000	7.574	2	0.000	33.324	−4.432	2.311
	6	1	1	0.788	2	0.938	0.476	2	0.062	5.917	−2.010	−0.038
	7	2	2	0.923	2	0.978	0.161	1	0.017	8.297	0.166	−0.926
	8	1	1	0.378	2	0.763	1.944	2	0.237	4.277	−1.929	−0.752
	9	2	2	0.383	2	0.735	1.920	1	0.265	3.958	−1.314	−1.143
	10	1	1	0.001	2	1.000	14.921	2	0.000	47.004	−5.660	2.465
	15	Ungrouped	2	0.262	2	0.811	2.680	1	0.140	6.192	0.194	0.318
	17	Ungrouped	2	0.336	2	0.543	2.179	1	0.457	2.523	−0.901	−0.194
	28	Ungrouped	1	0.147	2	0.848	3.840	2	0.143	7.398	−0.471	1.354
	60	Ungrouped	1	0.106	2	0.688	4.496	2	0.287	6.243	−0.236	1.168
	68	Ungrouped	1	0.081	2	0.948	5.024	2	0.044	11.175	−0.514	1.980
	89	Ungrouped	2	0.279	2	0.678	2.553	3	0.321	4.051	1.522	−0.666

.

Table 12. The results of classification.

		Classification	Predicted Group			Total
			1	2	3
Original	Count	1	35	0	0	35
		2	0	30	0	30
		3	0	0	27	27
		ungrouped	3	3	0	6
	N(%)	1	100.0	0	0	100.0
		2	0	100.0	0	100.0
		3	0	0	100.0	100.0
		ungrouped	50.0	50.0	0	100.0

shows the results of FDA. Among the six data, three of them were classified wrongly. These three data are 41.69, 39.59, and 45.00. All of them are supposed to be classified into the second category. However, 41.69 and 39.59 are classified into the first category initially, and 45.00 is classified into the third category. Meanwhile, the correct rate of discrimination is 100% except the six ungrouped data, and the correct rate of discrimination for all of the data is 96.9%.

Therefore, the 16 different working conditions are classified into three classes based on the value of the score of FDA. The range of the F-score for the first-class environment is 28.14–39.76. The range of the F-score for the second-class environment is 39.17–45.21. The range of the F-score for the third-class environment is 44.13–52.39.

3.4. Comparison Among Three Kinds of Environments

During the experiment, the shortest working time for the subjects was 30 min, so the values of the physiological and psychological indexes in the 30th minute were extracted to be further analyzed in order to make each data comparable.

3.4.1. Oral Temperature

The boxplot of oral temperature concerned with classification is shown in Figure 3. It can be observed that there are significant statistical differences among the three sets of data (p < 0.001). The value of oral temperature within the first-class environment is distributed in a much more concentrated manner than the second-class environment, but is more scattered than that of the third-class environment.

The maximum of the oral temperature of the subjects in the second-class environment increased by nearly 0.6 °C, and the minimum of it increased by nearly 0.2 °C to the first-class environment. The changing trend from the second-class environment to the third class is similar to the trend talked before. The maximum increased by nearly 1.5 °C and by nearly 1 °C for the increase of the minimum. Both of the increases for the latter are bigger than that for the former.

The average of oral temperature in the second environment is around 38 °C. This means that it almost reaches the safety physiological limit, and human health may suffer from the potential influence from heat stress. The average for the third-class environment is around 39 °C. This means that it almost reaches the tolerance physiological limit. In this case, if subjects continue working without resting, they will have a big chance to suffer from a thermal illness, and they cannot endure the physiological stress at this moment.

3.4.2. Heart Rate

The boxplot of heart rate concerned with classification is shown in Figure 4. Significant statistical differences were also found among the three sets of data (p < 0.001). The values of the heart rates of the subjects in the third-class environment are distributed in the most concentrated manner. The increase from the first-class environment to the second-class environment is obviously larger than that from the second-class environment to the third-class environment. Additionally, when the subjects were working in the second-class environment, the average values of the heart rates of the subjects were larger than 130 beats/min. When the subjects were working in the third-class environment; the average value of the heart rate was higher than 140 beats/min.

When the subjects are working in the second-class environment, the final values of the heart rates of the subjects maybe larger than 145 beats/min. The subjects may suffer from the potential influence that heat stress has. When the heart rate is higher than 140 beats/min, it is close to the safety physiological limit, 145 beats/min. It means that human health at this moment may suffer from the potential influence. Specifically, the data are the 30th minute for the subjects. If they continue working in this kind of environment, the heart rate may reach the tolerance physiological limit, and they will suffer from thermal illness.

3.4.3. Mean Skin Temperature

The boxplot of the mean skin temperature of the subjects concerned with classification is shown in Figure 5. It also can be observed that there are significant statistical differences among the three sets of data (p < 0.001). The general trend for both of the maximum and minimum of T_sk is increased from the first- to the third-class environment.

The safety physiological limit for the T_sk of the subjects was 37.7 °C [76]. The maximum of T_sk of the subjects for the first- and second-class environment were around and even lower than 37 °C, so the subjects working in these two kinds of environment had little chance to suffer from the influence that heat stress has. However, when the subjects were working in the third environment, the T_sk of some subjects already exceeded 37.7 °C when they worked for 30 min. This means T_sk has reached the safety physiological limit.

3.4.4. Feelings of Fatigue

Figure 6 shows the boxplot of FOF of the subjects concerned with classification. Significant statistical differences were also found among the three sets of data (p < 0.001). The general trend of FOF of the subjects is increase from the first-class environment to the third-class environment.

The data in Table 8 are from the subjects who were working in the 40 °C—RH90% environment, and some of the subjects even had nearly 20 fatigue symptoms. This means that their health had been seriously affected by the heat stress, and they cannot endure the heat stress at those moments. The feelings of fatigue among the subjects are important; they can judge their body state according to their feelings. However, feelings of fatigue are a subjective index; each subject may have a different result from each other even though they are in the same condition.

3.4.5. TSV

Figure 7 shows the boxplot of TSV of the subjects concerned with classification. TSV is also a subjective index for the subjects. The general trend for TSV is obviously increase from the first- to the third-class environment. Significant statistical differences were also found among the three sets of data (p < 0.001).

Most of the averages of TSV of the subjects for the first-class environment range from 1 to 1.7, and the average for the second-class ranges from 2 to 2.5. However, the average for the third-class ranges from 2 to 3. The result means that TSV increases with the increase in badness of the environment. Even though the subjects worked for 30 min, their value of TSV reached 3 in the environment, such as 40 °C—RH80% and 40 °C—RH90%.

4. Discussion

This study classifies the hyperthermal environment into three categories. The subjects’ performance can reflect the badness of the environment by observing the change in the physiological and subjective indexes. The badness of the environment increases from the first- to the third-class environment. Therefore, the classification can provide a better understanding of the hyperthermal environment under various temperatures and humidities. Additionally, the health of the person that works in this environment can be predicted based on his performance; then measures of precaution will be taken in time. Previous indexes mainly focused on physiological variables.

The influence of multiple physiological parameters should be considered comprehensively to evaluate human heat strain as accurately as possible. This study attempted to establish an evaluation index that comprehensively reflected human responses in hyperthermal environments. Therefore, in addition to the consensus reached within the field, the core temperature and HR should be considered. The human body must sweat a lot to maintain evaporation and heat dissipation in the thermal environment, which leads to frequent progressive dehydration. However, the thermal damage caused by dehydration is significant. Several studies have shown that dehydration leads to a decrease in plasma volume.

In addition, from the perspectives of physiology and medicine, T_or and HR belong to different physiological systems of the human body and have different physiological meanings. Therefore, in order to comprehensively evaluate human responses in hyperthermal environments, it is completely reasonable and meaningful to consider T_or and HR when assessing human heat strain. In addition, the reason why this study considered T_or instead of rectal temperature is that not only is T_or suitable for on-site measurement, but its measurement results are also easily accepted by subjects [6]. Finally, based on multiple human experimental data, this study comprehensively considered T_or, HR, and other parameters to establish a comprehensive score index for classifying hyperthermal environments.

Examples of these indexes include predicted 4 h sweat rate, operative temperature, skin wettedness, heat stress index, cumulative heat strain index (CHSI), perceptual strain index (PeSI), and physiological strain index (PSI). Meanwhile, a comprehensive heat stress index is proposed in this paper. It is composed of five variables: oral temperature, heart rate, mean skin temperature, feelings of fatigue, and TSV. These five variables include objective and subjective elements. It is not enough to evaluate the health of subjects only using the physiological indexes, subjective indexes are also important. The existence of human discrepancy would result in a wrong judgment. Therefore, the index F is composed of the two aspects to reflect the body state of the subjects much better. Three analysis method are used in this paper. The three methods are factor analysis (FA), clustering analysis (CA), and Fisher discriminant analysis (FDA). FA reflects a thought of dimensionality reduction, and gathers highly correlated variables. It can extract features that are easy to interpret, and reduce the number of variables that need to be analyzed and the complexity of problem analysis. CA and FDA are widely used, and the results from both of them are intuitive and concise. CA and FDA are easily used and the problems can be effectively handled by them.

The establishment of the environmental classification model in this study is based on statistical analysis methods. The statistical results show that the accuracy of the model is very high. In existing research, relevant studies have shown the effectiveness of statistical analysis methods [77]. In addition, according to the standardized equation of the model, T_a and RH are the two most important factors affecting human thermal response. They mainly determine the body’s heat dissipation through convection, conduction, and vaporization, and can be directly perceived through sensory perception. The literature has shown that wind speed can also affect human thermal response, but it usually needs to be combined with Ta to have a significant impact on the human body [78]. Therefore, studying the coupling effect between wind speed and Ta has practical significance, which is also one of our future research directions.

The single sample source, the small sample size, and the small number of environmental conditions are the limitations of this study. All of the subjects are male university students, whose physical quality differ from that of female subjects and male subjects from other districts. Further studies should examine larger samples with different genders and various occupations. Moreover, the subjects only work under heavy intensity, and the temperature and humidity in the experiment are only 32 °C–40 °C with an interval of 2 °C and 60%–90% with an interval of 10%. Meanwhile, the comprehensive index established in this study has not yet been applied and validated in practice. In future research, we will combine the comprehensive index with practical applications to improve and simplify the comprehensive index. Moreover, the temperature gradient of this study was set to 2 °C, which can meet the requirements of the comprehensive index in this study. However, in future research, it is necessary to further refine the temperature gradient setting of the experimental conditions to achieve precise classification and prediction of the environments. Furthermore, this study only considered temperature and humidity in the environmental conditions, which is also a limitation. In future research, other environmental conditions, such as wind speed and thermal radiation, should be added.

5. Conclusions

To evaluate the effects of hyperthermal environments on human responses accurately, a series of human heat strain experiments were conducted in a climate chamber in this study to obtain a large amount of human experimental data. Based on the dataset and reliable analyses, the following main conclusions were drawn:

(1) Considering the combined effects of human responses, a new comprehensive score index was developed based on the T_or, HR, and subjective parameters. The weightings of each parameter were determined using the principal component analysis method. Validation results showed that the new comprehensive score index could effectively and accurately classify different hyperthermal environmental conditions.

(2) Based on the obtained parameters of the subjects, two principal components with an explained variance of 92.131% were extracted by principal component analysis, and with the determination of weightings, sixteen different hyperthermal environmental conditions were classified into three categories based on the comprehensive score index. The F-score range for each type of environment was 28.14–39.76, 39.17–45.21, and 44.13–52.39, with the score increasing from the first type of environment to the third type of environment. The property of classification was examined by the analysis of physiological and subjective indexes. The three classes of environment are general, moderate, and heavy hyperthermal environment.

(3) For the first-level environmental conditions, the physiological and subjective indicators of the subjects were at their lowest at 30 min, which means that subjects working under these conditions have a low chance of experiencing heat stress. For the secondary environment, the physiological indicators of the subjects did not reach the safe physiological limit at 30 min, but they were already quite close. This means that if the subjects continue to work for a long time, they may be at risk of exposure to hot air. For the third type of environment, the values of various physiological and subjective indicators at 30 min were the highest among the three types of environments, and the values of physiological indicators exceeded the safe physiological limit.

This study combines physiological indicators and subjective parameters of the human body, and conducts research from multiple perspectives, including the cardiovascular system, temperature regulation system, thermal sensation, and fatigue level. It comprehensively and effectively investigates the degree and regularity of the effects of hyperthermal environments on human responses. A new comprehensive score index for all environmental conditions within the range of environmental parameters was ultimately established, providing a theoretical basis for the research and control management of safety issues for workers in different high humidity and heat environments. Based on the established comprehensive evaluation index, this research divides hyperthermal environments into three categories, which are based on the responses of the human body in each category. The environmental classification method can be well applied in high-temperature labor management. By classifying high-temperature labor environments, reasonable activity times can be established to ensure activity safety and improve efficiency. At the same time, it can also predict the physical condition of people in different environments and timely detect and avoid high-temperature risks.