Method for evaluating sound quality of rotary machine, ac machine drive sound control device, and electric power conversion device

Abstract

This method for evaluating the sound quality of a rotary machine (2) comprises: a step for selecting, as a value word of a language evaluation index relating to emotions caused by the drive sound of the rotary machine (2), an evaluation item selected from among evaluation items in an evaluation structure diagram created from results data deriving from performing a sensitivity investigation using an evaluation grid method relating to the results of an experiment involving listening to the drive sound; a step for selecting an evaluation item related to an impression element related to the sound quality or tone of a motor drive sound as an impression word of the language evaluation index with respect to the drive sound; a first evaluation step for evaluating the drive sound by using the value word; a second evaluation step for evaluating the drive sound by using the impression word; and a third evaluation step for evaluating the relationship among the three evaluation elements of the evaluation result obtained in the first evaluation step, the evaluation result obtained in the second evaluation step, and a physical feature quantity of the drive sound, and analyzing the result of the evaluation. By executing the two selection steps and the three evaluation steps, data is outputted as an evaluation result for the noisiness of the drive sound, which is indicated by the plurality of evaluation elements applied to the drive sound.

Description

Method for evaluating sound quality of rotating machine, AC machine drive sound control device, and power conversion device

[0001] The present disclosure relates to a method for evaluating the sound quality of a rotating machine, an AC machine drive sound control device, and a power conversion device.

[0002] Rotating machine control technology, which uses inverters to control the drive of motors, is widely used as a power source or power source for industrial and household equipment, and is often used near areas where people are active. In such cases, the various noises emitted by rotating machines can be unpleasant for users, and countermeasures must be taken. To do this, it is necessary to evaluate the drive noise emitted by rotating machines.

[0003] Noises emitted by office equipment used in an office environment, such as copiers or printers, are generated from multiple sources, including motors, paper, solenoids, etc., and cause discomfort to users. In order to reduce noise emitted from such office equipment that has noise sources that cause discomfort to users, it has been pointed out that in addition to quantitative evaluations such as sound pressure level, evaluation and improvement of sound quality, which is difficult to evaluate quantitatively, is necessary, and qualitative sound quality evaluations by humans have been performed (see, for example, Patent Document 1).

[0004] Another example is the use of sensory evaluation information from a user to determine whether the sound emitted from a vehicle is a pleasant sound for the user or to prevent the sound from causing discomfort to the user (see, for example, Patent Document 2).

[0005] Japanese Patent No. 4810245 Japanese Patent Application Laid-Open No. 2006-258510 Japanese Patent No. 3708292

[0006] Aures W. , “Acta Acoustica united with Acustica, Vol. 59, pp. 130-141, 1985 Junichiro Sanui, "Interview surveys for product planning: Current status and issues of traditional interview surveys and evaluation grid methods", Quality, 33(3), pp.13-20, 2003 Tomomi Takezawa, Takeshi Katahira, Yuna Kanyoshi, Masashi Sugimoto, Kazuo Shibuta, Noriko Nagata, Masataka Chiba, Kazuki Hamaoka, Megumi Fukatsu, Go Kataoka, "Stress structure during the COVID-19 pandemic and the alleviating effect of aromatherapy", Transactions of the Human Interface Society, 23(3), pp.337-348, 2021 Shinichiro Iwamiya, "Sensitivity of timbre in acoustic science series 1", "Evaluation and Creation of Timbre and Sound Quality," edited by the Acoustical Society of Japan, Corona Publishing, pp. 37-49, 2010. Kazuo Ueda, "Is there a hierarchical structure in timbre expressions?", Journal of the Acoustical Society of Japan, 44(2), pp. 102-107, 1988. Koji Abe, Kenji Ozawa, Yoichi Suzuki, Toshio Sone, "The influence of visual information on environmental sound perception," Journal of the Acoustical Society of Japan, 56(12), pp. 793-804, 2000. Daisuke Sawada et al., "Construction of a Sensory Evaluation Model for Smartphone Protective Films - Categorization Based on Individual Differences in Value Structure," Journal of the Japan Society of Kansei Engineering, Vol. 22, No. 2, pp. 207-216, 2023. Keizo Hori, "Determining the Number of Factors in Factor Analysis - Focusing on Parallel Analysis -", ​​Kagawa University Economic Review, 77(4), pp. 35-70, 2005. Takeshi Katahira et al., "Hierarchy of Evaluation in Measuring Sensibility Using SD Method - Focusing on the Diversity of Evaluation Factors in EPA Structure -", Journal of the Japan Society of Kansei Engineering, 17(4), pp. 453-463, 2018.

[0007] When evaluating these drive noises, the magnitude of the noise's harmonic components or frequency distribution is used. However, the way the noise is heard (auditory sensation) can vary greatly depending on the distribution, and it is not possible to evaluate these differences. As there is no established evaluation method that takes into account the audibility of motor drive noise, it is difficult to consider ways to reduce noise.

[0008] One approach is to create a sensory evaluation model that can evaluate the quality of sound based on human hearing, and then quantify and evaluate sound quality. For example, a technology for creating a sensory evaluation model for evaluating the sound quality of an image forming device is disclosed (see Patent Document 1). This model involves having subjects evaluate multiple types of sounds generated during image formation, performing factor analysis on the evaluation results, and performing covariance structure analysis on the acoustic physical quantities, factors, and unpleasantness of the sounds. Based on the analysis results, a model is created that explains the relationship between the acoustic physical quantities, one or more factors related to sound quality, and unpleasantness factors, as well as a path diagram that shows the relationship between the factors related to sound quality and the unpleasantness factors. The method disclosed in Patent Document 1 outputs a sound quality evaluation result as a single numerical value representing an overall unpleasantness level.

[0009] However, even sounds with the same low level of discomfort can be perceived as uplifting and pleasant, or as calming, and human hearing can be evaluated based on multiple emotions (hereinafter also referred to as affective value).If evaluation based on multiple affective values ​​like this can be performed, sound quality can be adjusted based on the evaluation results according to the application of the inverter-driven AC machine, enabling sound quality control that further enhances user comfort.

[0010] Furthermore, in the method disclosed in Patent Document 1, general acoustic evaluation indices are used as inputs to the sensory evaluation model. However, when targeting inverter-driven AC machines, due to the use of a switching method such as pulse width modulation (hereinafter also referred to as PWM, where PWM is an abbreviation for Pulse Width Modulation), the frequency distribution of the noise tends to be concentrated in a specific region; by taking such characteristics into consideration, it may be possible to construct a sensory evaluation model that more appropriately expresses human sensory value.

[0011] Therefore, the present disclosure aims to provide a method for evaluating the sound quality of a rotating machine, in which a sensory evaluation model is constructed that takes into account the characteristics of an inverter-driven rotating machine and appropriately expresses multiple sensory values.

[0012] The rotating machine sound quality evaluation method of the present disclosure includes a first selection step of conducting listening experiments on rotating machine drive sounds having different control methods in order to realize pleasant sound quality related to the auditory sensation of the rotating machine drive sound, conducting a sensory survey using an evaluation grid method based on the results of the listening experiments, and creating an evaluation structure diagram represented by a combination of nodes and paths based on the data obtained from the sensory survey, and selecting, from the evaluation items shown as nodes in the evaluation structure diagram, linguistic evaluation items selected based on the Katz centrality criterion as value words of a linguistic evaluation index related to emotions caused by the rotating machine drive sound; a second selection step of selecting, from the evaluation items shown in the evaluation structure diagram, linguistic evaluation items related to impression elements related to the sound quality or timbre of the motor drive sound as impression words of a linguistic evaluation index for the rotating machine drive sound; a first evaluation step of evaluating the pleasant sound quality of the rotating machine drive sound using the value words; and a second evaluation step of evaluating the pleasant sound quality of the rotating machine drive sound using the impression words. and a third evaluation step of evaluating and analyzing the relationship between three evaluation elements: the evaluation result obtained in the first evaluation step, the evaluation result obtained in the second evaluation step, and a feature quantity indicating the physical characteristics of the rotating machine drive sound. By executing the first and second selection steps and the first to third evaluation steps, an evaluation result is output that evaluates the pleasantness of the rotating machine drive sound indicated by the multiple evaluation elements for the rotating machine drive sound.

[0013] According to the method for evaluating the sound quality of a rotating machine disclosed herein, it is possible to provide a method for evaluating the sound quality of a rotating machine in which a sensory evaluation model is constructed that appropriately expresses multiple sensory values, taking into account the characteristics of an inverter-driven rotating machine.

[0014] 1 is a diagram showing a list of experimental stimuli used in the method for evaluating sound quality of a rotating machine of embodiment 1. FIG. 2 is a diagram for explaining an experimental environment used in the method for evaluating sound quality of a rotating machine of embodiment 1. FIG. 3 is a diagram for explaining an example of laddering used in the method for evaluating sound quality of a rotating machine of embodiment 1. FIG. 4 is an example of an evaluation structure diagram visualizing interview results of the evaluation grid method used in the method for evaluating sound quality of a rotating machine of embodiment 1. FIG. 5 is another example of an evaluation structure diagram visualizing interview results of the evaluation grid method used in the method for evaluating sound quality of a rotating machine of embodiment 1. FIG. 6 is a list diagram showing an example of evaluation items shown as evaluation word pairs regarding perceptual value used in the method for evaluating sound quality of a rotating machine of embodiment 1. FIG. 7 is a list diagram showing an example of evaluation items regarding impressions related to the sound quality and timbre of motor driving sound used in the method for evaluating sound quality of a rotating machine of embodiment 1. FIG. 8 is a diagram for explaining a specific example of the Semantic Differential (SD) method used in the method for evaluating sound quality of a rotating machine of embodiment 1. FIG. 9 is a diagram for explaining a specific example of a Likert scale used in the method for evaluating sound quality of a rotating machine of embodiment 1. FIG. 10 is a diagram for explaining factor loadings related to impressions (sound quality and timbre) of motor driving sound used in the method for evaluating sound quality of a rotating machine of embodiment 1. 13A is a diagram for explaining a multiple regression analysis method applied to the analysis of the sensory value of motor driving noise, used in the rotating machine sound quality evaluation method of embodiment 1. FIG. 13B is a diagram for explaining a hierarchical structure model used in the rotating machine sound quality evaluation method of embodiment 1, which examines whether the model structure is appropriate from the perspective of covariance structure. FIG. 13C is a diagram showing an example of a prediction model related to the pleasantness of motor driving noise, used in the rotating machine sound quality evaluation method of embodiment 1. FIG. 13D is a diagram for comparison with FIG. 13A. FIG. 13E is a block diagram showing the configuration of an AC machine driving noise control device according to embodiment 2. FIG. 13F is a block diagram showing the configuration of an AC machine driving noise control device according to embodiment 3. FIG. 13G is a block diagram showing the configuration of an AC machine driving noise evaluation unit in the AC machine driving noise control device of embodiments 2 and 3. FIG. 13H is a diagram showing an example of hardware of the AC machine driving noise control device of embodiments 2 and 3.

[0015] The method for evaluating the sound quality of a rotating machine according to this embodiment will be described below with reference to the drawings. In the drawings, the same reference numerals indicate the same or corresponding parts.

[0016] First Embodiment A method for evaluating the sound quality of a rotating machine according to a first embodiment will be described below with reference to the drawings.

[0017] <Selection Process> First, we will describe the procedure for creating a linguistic evaluation index suited to motor drive sounds by extracting an evaluation structure for the pleasantness of motor drive sounds. In order to comprehensively collect evaluation items to be used in creating the index, there is an interview method called semi-structured interview, in which general questions are determined in advance and more detailed questions are asked depending on the respondents' answers. The survey will be conducted using the evaluation grid method, which is a type of semi-structured interview. The evaluation grid method is said to be less dependent on the ability or technique of the researcher than regular interview surveys (see, for example, Non-Patent Document 2).

[0018] The 24 types of motor driving sounds shown in Figure 1 were used as experimental stimuli for the investigation. Here, the number of types of control parameters was set to 24, taking into consideration the way in which inverter control parameters are assigned (allocated).

[0019] Here, we will briefly explain each control method shown in Figure 1. Method 1 is based on a three-phase PWM method, and Method 2 is based on a two-phase PWM method. Method 3 is based on a modulation method that disperses the frequencies of harmonic components of the output line voltage using the method disclosed in Patent Document 3. The sound pressure presented as the experimental stimulus was set to 11 sones (approximately 65 dB) based on research on noise surveys. Here, a sone is a psychological unit that indicates the loudness of a sound perceived by the human ear, and one sone is defined as the loudness perceived by a human hearing a pure tone with a sound pressure level of 40 dB and a frequency of 1 kHz.

[0020] Research participants (also referred to as experimental participants or listeners; hereafter referred to as participants) wore headphones and listened to sound stimuli during the interview. As shown in Figure 2, the playback environment for the experimental stimuli consisted of a commercially available sound playback device (touch-panel PC), an audio amplifier, and headphones. Using the procedures of the evaluation grid method as a reference, the survey was conducted in three stages: categorizing the experimental stimuli based on similarity, ranking the stimulus sound groups, and conducting an interview using the evaluation grid method.

[0021] First, we will explain the classification of the experimental stimuli based on similarity in the first stage. Regarding the similarity of the auditory features of the experimental stimuli, they were classified into a small number of groups according to the procedures (a) to (d) shown below. Participants were allowed to check the experimental stimuli at any time during this classification. (a) Classify the cards corresponding to the 24 experimental stimuli into two groups based on auditory similarity. (b) Check whether the classified stimulus sound groups could be further classified into two groups. (c) If classification was possible, further classify them into two groups. (d) Repeat steps (a) to (c) until all stimulus sound groups could no longer be classified.

[0022] Specifically, the above steps (a) to (d) are as follows. First, in step (a), a subject listens to 24 types of sounds and determines that there is a similarity between sounds 1 to 12 and sounds 13 to 24. At this time, these sounds are classified into two groups: group A (1 to 12) and group B (13 to 24). Next, in step (b), it is checked again whether groups A and B can be further divided into "smaller groups." As a result, it is determined that group B can be divided into sounds 13 to 18 and sounds 19 to 24, respectively. In this case, in step (c), the two "small groups" divided in step (b) are classified into two groups: group C (13 to 18) and group D (19 to 24). Next, in step (d), steps (b) and (c) are repeated for groups C and D until classification is no longer possible.

[0023] Next, we will explain the ranking of the stimulus sound groups in the second stage. Participants ranked the stimulus sound groups classified in the first stage procedure based on the quality of the sound. Participants were asked to rearrange the stimulus sound groups in order of quality. They were then asked to rate the quality of the stimulus sound groups on a 7-point scale, with the best being given 7 points and the worst being given 1 point, as a relative evaluation.

[0024] The reason for using a seven-point scale is to enable participants to focus on differences in quality and make judgments. If the scale were broader than seven points, the differences would be more abstract, and if the scale were more detailed, participants would focus too much on specific differences, so seven points was chosen to strike a balance.

[0025] Furthermore, based on the above evaluation results, the experimental stimuli are organized into three groups based on the scores given based on the quality of the sound: a group of experimental stimuli belonging to the stimulus sound group with 1 to 2 points is classified as a low-rating group, a group of experimental stimuli with 3 to 5 points is classified as a medium-rating group, and a group of experimental stimuli with 6 to 7 points is classified as a high-rating group.

[0026] Finally, we will explain the interviews using the evaluation grid method in the third stage. The reasons and evaluation words for the comparative judgments that form the differences between the three groups created in the second stage above were elicited through interviews, and detailed elicitation (laddering) was conducted to determine the reasons for the comparative judgments, which will be explained in detail below.

[0027] The reasons for the comparative judgment were elicited by asking, "You say that this group is better than this group. Please list any reasons that come to mind for your judgment. Please note that the reasons do not necessarily have to apply to all of the groups you have classified; it is fine to list reasons that apply only to specific groups."

[0028] Laddering was carried out using two types of questions: laddering up and laddering down. An example of this laddering is shown in Figure 3. With laddering up questions (to elicit superordinate concepts), participants were asked to answer questions such as why they think the superordinate concept is good and what good things about it are. With laddering down questions (to elicit subordinate concepts), participants were asked to answer questions such as what the subordinate concept specifically means and in what respects they feel that way. When asking questions in terms of good / bad is not appropriate, a different way of asking can be used (for example, in the case of food, delicious / not delicious).

[0029] In this case, interviews were conducted using the evaluation grid method described above regarding both the positive aspects of sounds in the high rating group compared to sounds in the medium rating group (elements that should be noted in making sounds more comfortable), and the negative aspects of sounds in the low rating group compared to sounds in the medium rating group (elements that should be noted in reducing noise).

[0030] Figures 4 and 5 show the interview results of the three-stage evaluation grid method described above visualized as evaluation structure diagrams. Visualization using evaluation structure diagrams can be achieved using the evaluation structure visualization system (ESV system, where ESV stands for Evaluation Structure Visualization) described in Non-Patent Document 3, for example. This system visualizes the cognitive structure related to evaluation obtained using Sanui et al.'s evaluation grid method (see Non-Patent Document 2) as a graph of nodes and paths. The importance of these nodes is assigned based on their frequency of appearance in the responses, with a higher frequency being interpreted as a higher importance.

[0031] In the ESV system described above, words with the same meaning are integrated into a category to eliminate the influence of word variations. Within these integrated categories, terms (words) that could be interpreted as emotional factors or sound impressions are displayed as unhatched terms within a rectangular frame. On the other hand, hatched words within a rectangular frame are those that appear infrequently and were not integrated into a category, or those that do not correspond to either emotional factors or sound impressions. While the importance of the latter nodes is low, if the importance of related nodes is high, they are interpreted as being relatively important, and remain as displayed items, but do not contribute to the consideration of evaluation items.

[0032] Figure 4 shows an evaluation structure diagram for the positive aspects of motor drive noise, summarizing the interview results of all participants, and Figure 5 shows an evaluation structure diagram for the negative aspects of motor drive noise. Based on the information extracted through the laddering process, we created an evaluation structure diagram by placing superordinate items (hereafter referred to as superordinate items) on the left and subordinate items (hereafter referred to as subordinate items) on the right, connecting related items with lines. The superordinate items correspond to psychological value, and the subordinate items correspond to characteristics of motor drive noise, such as sound quality or timbre. Although responses (expressions) to the same subject or content may differ between participants, we standardized the expressions based on research on timbre expression shown in Non-Patent Documents 4, 5, and 6.

[0033] Each evaluation structure diagram shown in Figures 4 and 5 is a graph with nodes and paths, where the nodes correspond to concepts that represent how the subject is evaluated, and the paths represent relationships between concepts (for example, relationships in which multiple evaluation elements are linearly combined). In presenting the evaluation structure diagrams, evaluation items with high importance were selected based on the criterion of Katz centrality (for details on Katz centrality, see, for example, Non-Patent Document 7).

[0034] Generally, as the Katz centrality value decreases, the number of evaluation items increases, but the network becomes more complex and difficult to interpret; whereas, as the Katz centrality value increases, the number of evaluation items decreases, but the network becomes simpler and easier to interpret. In other words, by increasing the Katz centrality value, it is possible to visualize an evaluation structure consisting only of highly important nodes (see, for example, Non-Patent Document 7). Therefore, in Figures 4 and 5, the Katz centrality value was set to 0.1, which is a value that allows for both the variation of evaluation items and the interpretation of the evaluation structure diagram. Note that, in the analysis, the Katz centrality value can be adjusted to obtain an evaluation structure diagram with appropriate resolution.

[0035] The evaluation structure diagram is laid out so that nodes do not overlap, with nodes belonging to the value layer on the left and nodes belonging to the impression layer arranged (placed) from the middle to the right. In addition, the impression layer shows more specific sound quality or timbre of the motor drive sound as it moves to the right, and from right to left, the process of integrating auditory information about the motor drive sound, i.e., the process of value formation, is expressed.

[0036] The left-right relationship of the nodes reflects the hierarchy between concepts, but the distance in the left-right direction has no particular meaning.The properties of the nodes are represented by the type of border, with dashed and dotted lines representing emotional factors that form the positive aspects of motor drive noise, solid lines representing emotional factors that form the negative aspects of motor drive noise, and dotted lines representing tone and sound quality factors for motor drive noise.

[0037] <First Selection Step> Next, value words are selected from Figures 4 and 5. In Figures 4 and 5, if a line extends to the left from a certain item, the item at the end of the line is the higher-level value item. In other words, the item with no line extending to the left is the highest-level value item. From Figure 4, the highest-level values ​​that form the positive aspects are "calming" and "not tiring," and the lower-level value items that make up these are "not noisy," "relieving," "not unpleasant," "not bothersome," "easy to hear," "relaxing," and "can continue listening."

[0038] Furthermore, Figure 5 shows that the highest values ​​that make up the negative aspects are "unpleasant," "stressful," and "depressing," and the lower-level values ​​that make up these are "noisy," "unsettling," "anxious," "tiring," "worrisome," "rarely heard," "disgusting," and "dislike."

[0039] When the correspondence between these value items was compared with the evaluation word pairs expressing people's feelings toward general sounds reported in Non-Patent Document 6, correspondence between the evaluation items was found for many items except for "depressing" and "dislike." Based on this result, the nine evaluation word pairs shown in Figure 6 were selected as language indices for evaluating the value of motor drive noise. In other words, Figure 6 is a diagram showing a list of evaluation items that show evaluation items related to value as evaluation word pairs.

[0040] Note that Figure 6 does not include "doesn't tire you out," "can continue listening," and "tired" from the nodes that are evaluation items in Figures 4 and 5. The reason these evaluation items were excluded from Figure 6 is that the evaluation structure diagrams in Figures 4 and 5 summarize the interview results and extract the causal relationships between each concept, whereas Figure 6 differs from these diagrams in that it uses evaluation terms selected for use in hearing tests, and also because it was determined that factors other than the sound heard (participants) had a large influence on the formation of evaluations for the items listed as other nodes.

[0041] <Second Selection Step> Next, impression words are selected. Focusing on impression elements such as the sound quality and timbre of the motor drive sound in Figures 4 and 5, sound factors that create positive aspects include "dull sound," "low sound," "unpleasant sound," "appropriate pitch," "familiar sound," "clear sound," "quiet sound," "white noise," and "complex sound," while sound factors that create negative aspects include "sharp sound," "high sound," "unpleasant sound," "harsh sound," and "fluctuation sound."

[0042] In addition to these elements, a linguistic evaluation index for motor drive noise was created by combining evaluation words extracted using the evaluation grid method from literature on noise evaluation in the fields of general noise, home appliances, automobiles, and industrial equipment. As a result, the 25 evaluation items shown in Figure 7 were selected as linguistic evaluation indexes for the sound quality and timbre of motor drive noise. This concludes the process of selecting value words and impression words as linguistic evaluation indexes.

[0043] <Evaluation process> Next, in order to collect quantitative evaluation data of psychological reactions when listening to the motor drive sound, a subjective evaluation experiment was conducted using the created linguistic evaluation index. The experimental stimuli used in the evaluation were the 24 types shown in Figure 1, plus 24 types with a sound pressure of 20 [sone] (approximately 75 [dB]), for a total of 48 types of data. This is because sound pressure is generally a factor that cannot be ignored in noise evaluation, and 48 types (24 types x 2 sound pressure levels) were used in order to verify the effect of this sound pressure.

[0044] The experiment was conducted in the experimental environment shown in Figure 2, and participants performed subjective evaluations of each stimulus in the order of emotion (value) and impression (sound quality / timbre). Specifically, as shown in Figure 2, participants wore headphones and performed subjective evaluations by listening to the experimental stimuli, sounds emitted from a commercially available sound playback device (a touch-screen PC), through an audio amplifier. The diagram on the right shows an example of the operating screen used by participants in the experiment.

[0045] <First evaluation step> First, the experimental stimulus to be evaluated was listened to for 10 seconds in a resting state, and immediately thereafter, the experiment was conducted using a five-point SD method as shown in Figure 8 to answer the question "What emotions did you have when listening to the motor drive sound?", and responses (also called free responses) were obtained for the nine items related to value in Figure 6. Note that this SD method is a method commonly used to evaluate paired concepts such as those shown in Figure 6.

[0046] <Second evaluation step> Next, an experiment was conducted to obtain responses (also referred to as free responses) to the extent to which each of the 25 motor drive sound impression items in Fig. 7 applied to the subjects using a 5-point Likert scale as shown in Fig. 9. Here, since the sound impression does not have a corresponding concept, the impression evaluation items were evaluated using the Likert scale, which is used as a method of expressing the degree of application.

[0047] At this time, participants were allowed to listen to the experimental stimuli again at their discretion during the impression evaluation. The presentation order of the experimental stimuli was randomized and varied for each participant. Each stimulus was played for two seconds with a one-second interval between each, and each was evaluated once.

[0048] Here, whether or not participants could voluntarily listen again was determined based on the difference in psychological states when listening between value and impression. In other words, the value (emotion) evaluation aimed to measure the participants' emotions when listening to the motor drive sound. Emotions are greatly influenced by repeated listening, and participants are asked to answer about their own state immediately after listening, making it easier to answer than impressions, so participants were asked to answer without listening again. On the other hand, the impression evaluation aimed to measure the impressions obtained as participants' evaluation of the motor drive sound. Generally, repeated listening has little effect on the evaluation, and since there are a wide variety of evaluation items, confirmation was considered necessary, so relistening was made available at the participants' discretion. This concludes the evaluation process for the pleasant sound quality of rotating machine drive sounds using value words and impression words.

[0049] <Third Evaluation Step> First, in order to extract metrics (also called evaluation axes or evaluation scales) that exist behind the evaluation of the sound quality and timbre of the motor drive sound from the impression evaluation results of the subjective evaluation experiment, a factor analysis was performed on the impression evaluation item data. The factor analysis was performed using statistical analysis software JMP14Pro (JMP is a registered trademark of SAS Institute Inc.). The number of factors was determined by parallel analysis (see, for example, Non-Patent Document 8).

[0050] Here, maximum likelihood was used as the factor extraction method, and Promax rotation (oblique rotation) was used as the rotation method. When focusing on the actual state of human psychological evaluation axes, the Promax assumption (assuming that there is a correlation between factors) is in line with reality, so oblique rotation was used here. Rotation methods in factor analysis can be broadly divided into two: orthogonal rotation (Varimax) and oblique rotation (Promax). The difference is that orthogonal rotation extracts factors so that the correlation between factors is zero, while oblique rotation extracts factors assuming that there is a correlation between factors.

[0051] As a result of the factor analysis, three factors that form the impression of motor drive sound were extracted, as shown in Figure 10. In other words, Figure 10 is a table showing the factor loadings related to the impression (sound quality and timbre) of motor drive sound. Below, we will explain the content of each factor shown in Figure 10 and the correlation coefficients that represent the correlation between the factors.

[0052] First, for the first factor, items such as "pleasant," "beautiful," and "natural" had large positive values, while items such as "dirty" and "unpleasant" had large negative values. Because this factor is thought to reflect the aesthetic aspect of motor-driven sound, it was named "FI1: Aesthetic Factor." Next, for the second factor, items such as "powerful" and "powerful" had large positive values, while items such as "weak" and "small" had large negative values. Because this factor is thought to reflect the loudness of the sound, it was named "FI2: Power Factor." Finally, for the third factor, items such as "high" and "sharp" had large positive values, while items such as "low" and "dull" had large negative values. Because this factor is thought to reflect the metallic qualities of timbre, it was named "FI3: Metallic Factor."

[0053] Looking at the quantitative aspect of correlation between factors, the correlation coefficient between FI1: aesthetic factor and FI2: powerful factor was relatively large at -0.59, suggesting a relatively strong relationship between FI1: aesthetic factor and FI2: powerful factor in terms of the properties of motor drive noise. On the other hand, FI3: metallic factor had a correlation of -0.27 with FI1: aesthetic factor and 0.06 with FI2: powerful factor, indicating that these are independent impression evaluation results.

[0054] The value provided by motor drive sounds was categorized based on metrics (evaluation scales or evaluation axes) of impressions related to the sound quality and tone of motor drive sounds. In the analysis, a multiple regression analysis was performed using each of the nine value-related evaluation items as the objective variable and three impression-related factors as the explanatory variables, and the values ​​were categorized based on the similarity of the relationship between the obtained value and impressions. For the multiple regression analysis, a Stepwise variable selection method based on the AIC criterion (here, AIC is an abbreviation for Akaike's Information Criteria; the same applies hereinafter) was used.

[0055] An example of a specific result of multiple regression analysis is shown in Figure 11. For each evaluation item, the coefficient of determination (R2), which indicates the model accuracy of the sound quality evaluation model, and the partial regression coefficients (see the values ​​in the columns FI1 to FI3), which indicate the strength of the relationship between impression factors, are shown in a table in Figure 11. Note that the number of perceptual values ​​(hereinafter simply referred to as values) was set to three, but this is not limited to this.

[0056] Looking at the coefficients of determination shown in Figure 11, we see that they are generally high, indicating that the emotional value of motor drive sound can be expressed by sound quality and timbre. The coefficient of determination for "I often hear it" is small at 0.58, but this suggests that factors other than sound quality and timbre have a relatively large influence. In other words, this can be interpreted as a result of the strong influence of the participants' individual experiences other than sound. From this perspective, this was subsequently excluded from the sound quality evaluation model (hereinafter simply referred to as the model) (as with "tired").

[0057] Next, when we look at the structure of the partial regression coefficients, the value of motor drive noise can be broadly classified into three types. These differences in the relationship structure between value and impression mean that the nature of the value differs, and the pleasantness of motor drive noise has three aspects. Specifically, each value differs in the degree to which attention is paid to FI1: the aesthetic factor, which relates to the aesthetic evaluation of motor drive noise. The first value element is composed of "not loud," "pleasant," and "not bothersome," and since FI2: the powerful factor, which relates to the loudness or sharpness of the sound, and FI3: the metallic factor, contribute to its formation, it is thought to represent the value of low noisiness.

[0058] On the other hand, the second value element is composed of "calming," "relief," and "atmosphere," and represents a sense of security, a value that emphasizes the evaluative aspect of the first value element. This second value element shows no correlation with FI2: Power factor, but the partial regression coefficient of FI1: Aesthetic factor is large.

[0059] The third value element is composed of "I like it" and "It makes me feel at ease," and since a relationship was only found with FI1: Aesthetic factor, it is clear that it represents values ​​related specifically to taste.

[0060] Finally, by analyzing the relationship between the three elements of value that make up the pleasantness of motor drive noise, impressions regarding sound quality and timbre, and acoustic features as physical quantities, we build a model that can predict the pleasantness of motor drive noise from acoustic features. This model is also an appropriate sound quality evaluation model for expressing the pleasantness of motor drive noise.

[0061] The elements that form the value layers of the sound quality evaluation model are represented by the evaluation items "not noisy," "calming," and "like," which are three elements that were particularly selected from values ​​1 to 3 that are closely related to the impression of sound quality and timbre among the elements that make up the pleasantness of motor drive sound shown in Figure 11.

[0062] The impression layer is made up of three factors (FI1: aesthetic factor, FI2: power factor, FI3: metallic factor) related to the sound quality and tone of the motor drive sound extracted by the factor analysis in Figure 10. Note that for the value and impression factors, the average value of all participants is used as the representative value in order to correspond to the physical quantities.

[0063] Furthermore, in addition to the five typical sound quality evaluation indices (Loudness, Sharpness, Roughness, Fluctuation Strength, and Tonality), the physical layer elements selected were "center frequency," "kurtosis," "frequency spread," "entropy," and "major code ratio" as features that reflect the spectral bandwidth characteristics (also called frequency characteristics).

[0064] In constructing the model, it was assumed that higher layers are influenced by lower layers, based on the hierarchical nature of sensibility. Under this assumption, a multiple regression analysis was performed on all elements of the upper and intermediate layers, with each element of the upper layer as the objective variable and each element of the lower layer as the explanatory variable. Here, a variable selection method (stepwise method) based on AIC (Akaike's Information Criteria) was employed for the multiple regression analysis. This procedure makes it possible to construct an optimal model that minimizes the error between observed values ​​and predicted values ​​by the model.

[0065] Since the AIC criterion is used, this variable selection method searches for a combination of acoustic features that is optimized to reflect the magnitude of the main influence with the minimum number of parameters in explaining the model output variable (here, value or impression).

[0066] As shown in FIG. 13A , which will be explained in detail below, the result is an optimal model structure from the perspective of the above-mentioned variable selection method. In the process of variable search, a comparison was made between cases where band features were included and cases where they were not included. In this result, a better model was selected when band features were included than when they were not included, which indicates that the model's consideration of spectral band features contributed to improved accuracy.

[0067] Furthermore, we supplementarily examined whether the model structure was appropriate from the perspective of covariance structure. Specifically, we applied covariance structure analysis to the hierarchical structure model obtained by multiple regression analysis and evaluated its appropriateness from the perspective of goodness of fit. For the multiple regression analysis, we used the statistical analysis software JMP14Pro mentioned above, and for the covariance structure analysis, we used the lavaan package running in R.

[0068] The constructed hierarchical structure model is shown in Figure 12. This model visualizes the results obtained from the multiple regression analysis, using arrows to indicate the elements and their relationships. The three-layer structure shown in Figure 12 was adopted for the pleasant sound prediction model because it suggests that a hierarchy exists in sensibility, a notion that is widely accepted in the field of sensibility engineering, and also because it is a prerequisite for sensibility modeling that "it is desirable to reproduce the sensibility structure as closely as possible" (see Non-Patent Document 9).

[0069] The numbers in the figure indicate that the element at the base of the arrow influences the element at the tip of the arrow, and the numbers near the arrows represent partial regression coefficients. The line type and thickness of the arrows visualize the information on the partial regression coefficients, and the line type indicates the direction of the influence (solid line: positive influence, dotted line: negative influence). The thickness of the arrows also indicates the strength of the influence. Specifically, as shown in Figure 12, for example, "like," value 3, only has a thick solid line indicating a positive influence from aesthetic factors (value value is 0.93), and no dotted arrow indicating a negative influence. This is because the path (dotted arrow) was deleted due to the small influence found in the results of covariance structure analysis.

[0070] In path analysis of covariance structure analysis, the existence of a path (arrow) means that the value of the regression coefficient in the regression model is different from zero, i.e., there is an influence (the presence or absence of an influence can be determined by testing whether the value of the regression coefficient is significantly different from zero). The absence of a path means that there is no influence from the perspective of regression analysis, i.e., the regression coefficient is determined not to be significantly different from zero. From the above, it can be said that it is suggested that "liking" is formed solely by aesthetic factors.

[0071] First, we focus on the accuracy of the model. As a result of the covariance structure analysis, the goodness of fit of the model was GFI (Goodness of Fit Index) = 0.998 and AGFI (Adjusted Goodness of Fit Index) = 0.995, and the model was evaluated as being a good fit to the observed data.

[0072] The accuracy of prediction from acoustic features for each element of value and impression was also evaluated in terms of coefficient of determination. The coefficients of determination for FI1 (aesthetic factor), FI2 (power factor), and FI3 (metallic factor) were 0.86, 0.95, and 0.83, respectively, demonstrating that the acoustic features can be used to evaluate the impression of sound quality and timbre of motor drive sounds with good accuracy. Furthermore, it was found that value could also be predicted with generally good accuracy, with "not noisy" having a coefficient of 0.88, "calming" having a coefficient of 0.81, and "like" having a coefficient of 0.77.

[0073] Next, we focus on the model structure. Multiple regression analysis revealed that the acoustic features that contribute to the formation of value and impression are loudness, tonality, kurtosis, and center frequency. Furthermore, focusing on the relationship between impressions and acoustic features, we find that loudness contributes to the formation of FI1 (aesthetic factor) and FI2 (power factor), while tonality contributes to the formation of FI3 (metallic factor). This corresponds well with findings from noise research. Another study found that features reflecting spectral band characteristics are also necessary information for discriminating the impression of motor drive sounds. Strictly speaking, the term "tonality" is a general term, and there is also the tonal audibility (TA) standard defined in the international standard IEC 61400-11:20, which is classified as one type of tonality. However, we used the tonal audibility standard here.

[0074] The reason why I stated earlier that this corresponds to findings from noise research is as follows. Specifically, when focusing on the "not loud" aspect of the model, there are two factors that reduce this: the "power factor" and the "metallicity factor." "Metallicity" is generally interpreted as the pitch of a sound, and it is well known that both loudness and sharpness have the effect of reducing the pleasantness of pleasant awakening sounds. This corresponds to the structure of the "power factor" and "metallicity factor" mentioned above. On the other hand, it is known that "metallicity" is a complex impression, and in addition to the pitch of a sound represented by sharpness, the degree of prominence of the tone reflected in tonality (pure tonality) is also thought to influence its formation. It is also well known that the level of pure tonality affects "noisiness." For these reasons, I stated that this corresponds to findings from noise research.

[0075] Finally, the accuracy of the constructed model was confirmed. Fig. 13A shows the correspondence between the predicted and measured values ​​of the constructed model and the coefficient of determination, while Fig. 13B shows the correspondence between the predicted and measured values ​​of the Aures model, a conventional method described in Non-Patent Document 1, and the coefficient of determination. The coefficient of determination in Fig. 13B is 0.78, while the coefficient of determination in Fig. 13A is 0.88, indicating that the pleasantness of motor drive sound can be captured with higher accuracy than the conventional pleasantness model.

[0076] As explained above, the model constructed in the method for evaluating sound quality of a rotating machine according to the first embodiment is a good fit to the data obtained from the experimental stimuli, and it has been found that the acoustic features can be used to explain psychological responses to the pleasantness of motor drive sounds with a good degree of accuracy.

[0077] Second Embodiment Next, an example in which the sound quality evaluation method for a rotating machine according to the first embodiment is applied to a specific device will be described below with reference to the drawings. FIG. 14 is a block diagram showing the configuration of an AC machine drive noise control device 1 according to the second embodiment. The AC machine drive noise control device 1 includes a power converter 11, a switching command generation unit 12, an AC machine drive noise evaluation unit 13, and a drive noise measurement unit 14. Here, the switching target may be, for example, an inverter-driven motor. The drive noise measurement unit 14 is an arbitrary sound collector that collects motor drive noise, measures motor drive noise Sd (hereinafter also referred to as AC machine drive noise Sd), and outputs the measured noise to the AC machine drive noise evaluation unit 13. FIG. 16 is a block diagram showing the configuration of the AC machine drive noise evaluation unit 13 according to the second embodiment. The AC machine drive noise evaluation unit 13 includes a sound quality evaluation unit 31 that performs sound quality evaluation processing for an AC machine using the AC machine drive noise evaluation model created according to the procedure of the first embodiment, a sound quality evaluation index calculation unit 32 that calculates loudness and tonality from the motor drive noise Sd, and a frequency component calculation unit 33 that calculates kurtosis and center frequency.

[0078] The sound quality evaluation unit 31 calculates and outputs value evaluation values ​​for three values, "not noisy," "calming," and "likeable," from the input values ​​of loudness, tonality, kurtosis, and center frequency.

[0079] Here, a value evaluation value and a value command value, which is an evaluation target value, are input to the switching command generation unit 12. The switching command generation unit 12 generates and outputs a switching command Gt by calculation such that the value evaluation value coincides with the value command value.

[0080] Here, the value command value may be appropriately selected from the three values ​​of "not noisy," "calming," and "likeable" depending on the application to which the AC machine drive sound control device 1 is applied. For example, when applied to elevators, noise reduction is important, so weighting is performed to maximize the value of "not noisy," while when applied to air conditioners, which are installed in close proximity to people's living spaces, weighting is performed to maximize the value of "likeable."

[0081] A switching command Gt is input to the power converter 11 from the switching command generating unit 12, and the power converter 11 performs switching of the power module in accordance with the switching command Gt to supply power to the AC machine.

[0082] According to the AC machine drive sound control device 1 of embodiment 2, it is possible to drive the AC machine 2 (also called the rotating machine 2) with a motor drive sound having a value evaluation value according to the value command values ​​of the three values ​​desired by the user: "not noisy," "calming," and "likeable," which leads to an improvement in the auditory value of the user.

[0083] Third Embodiment Next, another example in which the rotating machine sound quality evaluation method of the first embodiment is applied to a specific device will be described below with reference to the drawings. FIG. 15 is a block diagram showing the configuration of an AC machine drive noise control device 1a according to the third embodiment. Compared to the AC machine drive noise control device 1, the AC machine drive noise control device 1a does not have the drive noise meter 14, but instead newly includes a current meter 15 and an AC machine drive noise estimator 16. The AC machine drive noise estimator 16 calculates and outputs an AC machine drive noise estimate Sdest from the switching command Gt and the measured current values ​​Ius, Ivs, and Iws. The AC machine drive noise estimate Sdest is input to the AC machine drive noise evaluator 13 instead of the AC machine drive noise Sd. The AC machine drive noise control device 1a uses the AC machine drive noise estimate Sdest to perform the same processing as the AC machine drive noise control device 1 and control the AC machine 2.

[0084] In addition to the effects of embodiment 2, the AC machine drive noise control device 1a of embodiment 3 has the effect of eliminating the need for a motor drive noise measuring device (also called a drive noise measuring instrument), making it possible to add a control mechanism that improves the sound quality of the motor drive noise even in existing equipment where it is difficult to install a motor drive noise measuring instrument.

[0085] Here, by applying the AC machine drive noise control device 1 of embodiment 2 or the AC machine drive noise control device 1a of embodiment 3, it is possible to manufacture a power conversion device that has the AC machine drive noise control device 1 of embodiment 2 or the AC machine drive noise control device 1a of embodiment 3, and that inputs the switching command Gt from the switching command generation unit of each AC machine drive noise control device to the power converter of each AC machine drive noise control device, thereby converting the power input to the AC machine 2 to be controlled.

[0086] Each of the hardware components of the AC machine drive noise control device 1 and the AC machine drive noise control device 1a, including the switching command generation unit 12, the AC machine drive noise evaluation unit 13, the AC machine drive noise estimation unit 16, the sound quality evaluation unit 31, the sound quality evaluation index calculation unit 32, and the frequency component calculation unit 33, is configured with a processor 100 and a storage device 101, as shown in FIG. 17 . The storage device 101 includes a volatile storage device such as a random access memory (not shown) and a non-volatile auxiliary storage device such as a flash memory. Alternatively, a hard disk auxiliary storage device may be included instead of the flash memory. The processor 100 executes a program input from the storage device 101. In this case, the program is input to the processor 100 from the auxiliary storage device via the volatile storage device. The processor 100 may output data such as calculation results to the volatile storage device of the storage device 101, or may store the data in the auxiliary storage device via the volatile storage device.

[0087] Although various exemplary embodiments and examples are described in this disclosure, the various features, aspects, and functions described in one or more embodiments are not limited to the application of a particular embodiment, but may be applied to the embodiments alone or in various combinations. Therefore, countless variations not illustrated are anticipated within the scope of the technology disclosed in this specification. For example, this includes cases where at least one component is modified, added, or omitted, or where at least one component is extracted and combined with components of another embodiment.

[0088] REFERENCE SIGNS LIST 1, 1a AC machine drive noise control device, 2 AC machine (rotating machine), 11 power converter, 12 switching command generation unit, 13 AC machine drive noise evaluation unit, 14 drive noise measurement device, 15 current measurement device, 16 AC machine drive noise estimation unit, 31 sound quality evaluation unit, 32 sound quality evaluation index calculation unit, 33 frequency component calculation unit, 100 processor, 101 storage device

Claims

1. a first selection step of conducting listening experiments on rotating machine drive sounds with different control methods in order to realize pleasant sound quality related to the auditory sensation of the rotating machine drive sounds, conducting a sensory survey using an evaluation grid method based on the results of the listening experiments, creating an evaluation structure diagram represented by a combination of nodes and paths based on the data obtained from the sensory survey, and selecting, from the evaluation items shown as nodes in the evaluation structure diagram, evaluation items in language selected based on the Katz centrality criterion, as value words of a linguistic evaluation index related to emotions caused by the rotating machine drive sounds; a second selection step of selecting, from the evaluation items shown in the evaluation structure diagram, evaluation items in language that relate to impression elements related to the sound quality or timbre of the motor drive sound as impression words of the language evaluation index for the rotating machine drive sound; a first evaluation step of evaluating the rotary machine drive noise in terms of sound comfortability using the value words; a second evaluation step of evaluating the pleasantness of the rotary machine drive noise using the impression words; a third evaluation step of evaluating and analyzing the relationship between three evaluation elements, namely, the evaluation result obtained in the first evaluation step, the evaluation result obtained in the second evaluation step, and a feature quantity indicating a physical feature of the rotating machine drive sound; and A sound quality evaluation method for a rotating machine, characterized in that by executing the first and second selection steps and the first to third evaluation steps, an evaluation result evaluating the pleasantness of the rotating machine drive sound indicated by a plurality of evaluation elements for the rotating machine drive sound is output.

2. 2. The method for evaluating sound quality of a rotating machine according to claim 1, wherein the feature quantities indicating the physical characteristics of the rotating machine driving sound are represented by acoustic feature quantities and frequency characteristics.

3. 2. The sound quality evaluation method for a rotating machine according to claim 1, wherein the value words are selected and used by an evaluation grid method based on evaluation data obtained from free responses of listeners when listening to the rotating machine drive sound.

4. 2. The sound quality evaluation method for a rotating machine according to claim 1, wherein the impression words include evaluation items in language selected using an evaluation grid method based on evaluation data obtained from free responses of listeners when listening to the rotating machine drive sound.

5. 2. The method for evaluating sound quality of a rotating machine according to claim 1, wherein the third evaluation step is a step of creating a path diagram that linearly combines the three evaluation elements.

6. a path is connected in the order of the feature amount of the rotating machine driving sound, the evaluation result obtained in the first evaluation step, and the evaluation result obtained in the second evaluation step, thereby forming a three-stage hierarchical structure; 6. The method for evaluating the sound quality of a rotating machine according to claim 1, wherein a feature quantity of the rotating machine drive sound and the evaluation result obtained in the first evaluation step, and the evaluation result obtained in the first evaluation step and the evaluation result obtained in the second evaluation step, are linearly combined to create a sound quality evaluation model of the rotating machine drive sound that inputs the feature quantity of the rotating machine drive sound and outputs a numerical value that indicates an affective value arising from auditory sensation.

7. a sound quality evaluation unit that performs sound quality evaluation processing of rotary machine driving noise using the sound quality evaluation model of rotary machine driving noise according to claim 6; a drive noise measuring device that measures the drive noise of the AC machine and outputs a measurement value of the drive noise of the AC machine; an AC machine drive noise evaluation unit that outputs a numerical value indicating the perceptual value for the drive noise of the AC machine as a value evaluation value based on the output of the sound quality evaluation unit and a measurement value of the drive noise of the AC machine; a switching command generation unit that generates a switching command based on a value command value that is a preset evaluation target value for the driving noise of the AC machine and the value evaluation value; An AC machine drive noise control device comprising:

8. a sound quality evaluation unit that performs sound quality evaluation processing of rotary machine driving noise using the sound quality evaluation model of rotary machine driving noise according to claim 6; an AC machine driving noise estimation unit that estimates the driving noise of the AC machine based on the voltage or current of the AC machine and outputs the estimated value of the driving noise of the AC machine; an AC machine drive noise evaluation unit that outputs a value evaluation value that indicates the perceptual value for the AC machine drive noise estimated value based on an output of the sound quality evaluation unit; a switching command generation unit that generates a switching command based on a value command value that is a preset evaluation target value for the driving noise of the AC machine and the value evaluation value; An AC machine drive noise control device comprising:

9. 8. A power conversion device comprising the AC machine drive noise control device according to claim 7, wherein a switching command from the switching command generation unit of the AC machine drive noise control device is input to a power converter of the AC machine drive noise control device to convert power input to the AC machine to be controlled.

10. A power conversion device having the AC machine drive noise control device described in claim 8, characterized in that a switching command from the switching command generation unit possessed by the AC machine drive noise control device is input to a power converter possessed by the AC machine drive noise control device, thereby converting the power input to the AC machine to be controlled.