3.2. Outcome of the Hand Wash Action Separation Analysis
The threshold flow rate values were set so that a 99.5% confidence interval could be obtained from the toilet usage data in all of the measured rooms. The aforementioned statistical values for full and half toilet flushes are shown in
Table 3 (The upper limit is the average value minus the value of 3σ, the lower limit is the average value minus the value of σ. The toilets used in this investigation were tank-type toilets, so the tank filling time per flush statistic in
Table 3 shows the amount of time it takes for new water to be supplied to the tank reservoir after each type of flush.).
However, in most cases the toilet use action is expected to also include a hand washing action. Although the values in
Table 3 only reflect the flow volume and time of toilet flushes in particular, in practice the flow volume and time of hand washes also count towards the actual water usage.
Therefore an attempt was made to obtain measurements of both flow and time values in regards to the washing of hands after a toilet usage action. In regards to hand washes (
n = 60), 15.03 +/− 7.99 s was determined to be the average necessary time for the completion of the action. Moreover, the time required to start the hand wash after finishing the actual toilet usage action is 3.5 +/− 0.99 s. It is important to consider both this time gap as well as the duration of the actual action for determining the upper value of the toilet water supply if the values for hand washes are to be added to the total. Hence, the upper value is set at 46 s, as calculated by taking the sum of the 3σ values for both full and half flushes in order to reach a 95% confidence interval (15.03 + 7.99 × 3 + 3.50 + 3 × 0.99 = 45.47). This value is bigger than the water flow time threshold value as seen in
Table 3, thus if hand washing is taken into consideration the upper limit of water flow time can be set at 46 s regardless of the type of flush. As for the lower flow time limit, this was set as the average of each measured room’s −3σ, assuming no hand washing action takes place.
Furthermore, the average cold water amount consumed for the washing of hands was determined to be 0.8 +/− 0.62 (L), whereas the average hot water consumption was 0.8 +/− 0.61 (L) according to the results of the investigation. Thus, the model of water flow volume and flow time as measured for a toilet use action is presented in
Table 4.
3.3. An Example of Setting Linear Discriminant Function
As mentioned in
Section 2.4, a linear discriminant function was used for each of the six investigated guest rooms in order to separate full and half flushes during the toilet activities and the showering and filling of the bathtub during the bathing activities. An example of the linear discriminant function as used for the toilet activities is shown below.
For the toilet usage total water volume and total time, both of which have very similar values during both full and half flushes (as can be seen in
Table 4), are taken into account. The discriminant analysis is conducted in order to properly identify each type of flush, thus in other words to avoid erroneously determining what kind of flush was used.
In case the variable for the water volume is set as “
x1” and the one for time as “
x2”, and given the probability distribution of
x = [
x1,
x2]′ in the population [
1], then the following Equation (1) can be assumed to represent the normal distribution N (μ
[1], Σ):
On the other hand, the following Equation (2) is assumed to represent the normal distribution N (μ
[2], Σ) given probability distribution of
x = [
x1,
x2]′ in the population [
2] and whereby Σ is considered to be the same as above:
If there would be a sample that belongs to either population [
1] or [
2], but for which it the designation is not known, then the correct population can be determined if the values of
x1 and
x2 are given. For this purpose the squared Mahalanobis distance is defined as follows:
whereby the Σ
−1 (with the superscript indicating the inverse matrix) is defined as:
Equation (3) represents the extension of the one dimensional Mahalanobis *1 equation to two dimensions.
Note *1 The square of the Mahalanobis distance is the distance from the sample’s x1 to each population, and can be defined as follows: .
D[k]2 is a quantitative measure of the statistical distance from x to the population mean vector μ[k] of the population [k]. If , representing a unit matrix, then D[k] would be an Euclidean distance. However generally , so in that case D[k] represents a value obtained by the adjustment of the Euclidean distance by variance and covariance. In case the two-dimensional normal distribution N (μ, Σ) is given by the probability density function *2, which when transformed becomes like Equation (5), then the corresponding relationship between the probability density function and the squared Mahalanobis distance is D2 = (x − μ)’Σ−1 (x − μ).
Note *2 The joint-probability density function of the two-dimensional normal distribution is generally given as:
The discriminant is set using the Mahalanobis distance as follows:
Here, we can use Equation (7), which is based on Equation (3).
However, one should consider that . Equation (8) divided by 2 is called linear discriminant function, whose discriminant is determined as in Equation (9).
With full and half flushes designated as populations [
1] and [
2] respectively, the discriminants for each measured target room were calculated.
Table 5 provides an example of results obtained in the examination of toilet activities in room A.
Given the full toilet flush data (based on the population [
1], as defined in Equation (9)) we can calculate the following:
On the other hand, from the half toilet flush data (based on population [
2], as defined in Equation (9)) we get the following:
As a result we get the following equation:
Thus, as explained earlier, the equation for the estimation of the linear discriminant function according to Equation (8) is the following:
The following is the discriminant equation for the values of the water flow (
x1) and the time (
x2) as shown in
Table 5:
Note that with the same calculation the coefficients for each of the investigated guest rooms can be set as shown in
Table 6.
If the above equation are used in order to calculate the discriminant values of the different toilet activities, then the identification rate is 100%.
3.4. Model of Toilet and Shower Actions
The toilet usage frequency (both full and half flushes), as well as the consumption, time and hot water temperature data, as measured in the six guest rooms, are shown per person and per room in
Table 7.
The output values gained from the activity separation algorithm show the toilet and shower uses per room. The toilet use and shower use per person were then calculated by dividing the output value by the average number of occupants.
Data from toilet use revealed that full flushes occur more than half flushes in the hotel environment compared to the mean Vietnamese household, although the total flush time is similar when compared to the results from the earlier study (
Table 8)
This may be due to the difference in users’ economic rationality. The toilet usage model in the previous study was targeted towards general households, so the burden of expenses was lowered by saving water. However, at places such as hotels, users have to pay a fixed amount of money, irrespective of the actual amount of water used. Therefore, their water-saving awareness level is considered low in the context of this study.
The showering data was compared with the results of an earlier study, in which the showering habits in Japan were measured and explored by means of a questionnaire. The previous study’s results [
10] are shown in
Table 9.
The research data from the Vietnamese hotel indicated a decrease in water usage, showering time and temperature of hot water when compared to the average Japanese household. Comparison of the instantaneous flow, the flow amount divided by the flow time, shows 6.6 L/min in the Vietnamese hotel to 10 L/min in the Japanese model. The variation in values between Vietnamese and Japanese data is caused by the different characteristics of the shower models used, not because of differences between the countries themselves. The average shower time in Vietnam is 7.3 min/use, which is close to the values seen in the Japanese summer data. This study shows that results obtained in Ho Chi Minh City, which has a consistent climate all year round, strongly resemble the data obtained in Japan during the summer. Moreover, it is believed that the difference seen in hot water temperatures is caused by the difference in climate zones of the investigative sites; namely the Torrid Zone in Ho Chi Minh and the Temperate Zone in Japan.