JP3770232B2 - Waveform inspection device - Google Patents

Description

【００３５】
決定係数ｒ^２は、回帰直線の適合度を示し、０以上１以下の値をとる。そして、ｒ^２＝１が完全な適合を意味する。従って、図６（ａ）に示す良品のＦＦＴ波形のように、周波数が増加するにつれてパワースペクトルがほぼ直線状に下降している場合には、ｒ^２は１に近い値をとり、図６（ｂ）に示す不良品のＦＦＴ波形のように、歪んだ箇所があると、ｒ^２は１から離れた値となる。
【００３６】
そこで、ステップ３で求めた良品と不良品のｒ^２が、両者を弁別できるだけ差があるか否かを判断する（ＳＴ４）。そして、弁別できるほど差がない場合には、ステップ２に戻り、βの値を変更し、前後幅の再設定を行う。
【００３７】
そして、ステップ４の分岐判断でＹｅｓとなると、求めた決定係数ｒ^２を特徴量値に変換する（ＳＴ５）。つまり、上記したように、決定係数は、良品の方が１に近い大きい値をとるが、本実施の形態では、他の特徴量（説明を省略）との関係から良品の場合の特徴量値の数値が小さく、不良品の場合が大きくなるようにしているので、それにあわせるべく決定係数を下記の変換式により特徴量値に変換する。
特徴量値＝１０００−（ｒ^２×１０００）
そして、求めた特徴量を出力する（ＳＴ６）。図６のＦＦＴ波形の場合の特徴量としては、例えば図７に示すようになり、良品と不良品とを確実に弁別できる。
【００３８】
なお、図５に示すフローチャートの各処理ステップのうち、ステップ５における変換は、上記したものに限ることはなく、例えば、単に決定係数ｒ^２に所定の定数を乗算したりする他、各種の変換を行うことが可能となる。さらに、決定係数をそのまま用いる場合には、ステップ５の処理自体をなくしてもよい。
【００３９】
また、ステップ４の分岐判断は、学習機能の時に実行され、調整によりステップ２で行った前後幅が決定され、実際の検査時においては、中心周波数や前後幅を初期設定値として持ち、特徴量演算部１５における実際の処理は、ステップ３，５，６を実行するようになる。
【００４０】
判定部１６は、特徴量演算部１５から出力された特徴量に基づいて判定処理を行うもので、例えば閾値処理により良否判定を行うことができる。また、特徴量が複数種出力される場合には、それらを総合的に判断することももちろんできる。
【００４１】
特徴量演算部１５の処理アルゴリズムとしては、上記した第１の実施の形態のものに限ることはなく、各種の変更実施が可能である。例えば、第２の実施の形態である図８に示すフローチャートを実施する機能にすることができる。このフローチャートも、学習機能を含んだものである。
【００４２】
まず、取得したＦＦＴ波形を微分する（ＳＴ１１）。すなわち、上記したように良品の場合のＦＦＴ波形は、ほぼ直線状となっているため、変形量のばらつきは小さい。これに対し、不良品の場合のＦＦＴ波形は、歪み・窪みがあるため、変化量のばらつきが大きくなる。そこで、微分値に基づいて特徴量を抽出するようにした。そして、係る処理を実行することにより、例えば、図６（ａ），（ｂ）に示すＦＦＴ波形は、それぞれ図９（ａ），（ｂ）に示すようになる。この図９における横軸は周波数、縦軸はスピーカの音再生のスペクトルパワーの微分値である。
【００４３】
次に、取得したＦＦＴ波形の微分波形に対する中心周波数を設定する（ＳＴ１２）。この中心周波数も、上記した実施の形態と同様に、入力装置を操作してパラメータとして設定する。つまり、予め歪みが生じる周波数がわかっている場合には、その値を中心周波数に設定することができる。なお、本実施の形態では、中心周波数を９００Ｈｚに設定している。また、本実施の形態では、不良品の場合にＦＦＴ波形が下降している区間に対して歪みが発生するので、例えば図９に示した微分波形を表示装置に出力させるとともに、それをみて全体として微分値が負になっている区間における適当な位置を中心周波数に決定する。そして、決定した値を数値で入力するようにしてもよいし、図示するように縦軸方向に伸びるガイドバーを重ねて表示し、入力装置を操作してガイドバーを左右に移動し、所定位置で決定することにより、中心周波数を設定することもできる。
【００４４】
次に、計算に使用するデータ数を決定する（ＳＴ１３）。本実施の形態では、中心周波数ｆ１に対して、前後等幅としているため、入力装置を介して「γ」を代入し、これを受けて「ｆ１−γ」と「ｆ１＋γ」を算出することにより、前後幅を決定する。このとき入力するγが調整パラメータとなる。
【００４５】
そして、「ｆ１−γ」〜「ｆ１＋γ」の対象範囲におけるピーク−ピーク値を求める（ＳＴ１４）。すなわち、上記したように良品の場合のＦＦＴ波形の微分値のばらつきは小さいが、不良品の場合のＦＦＴ波形の微分値は、歪み・窪み部分で正負が生じ、変化量のばらつきが大きくなる。従って、ピーク−ピーク値は、良品ほど小さく、不良品ほど大きくなる。
【００４６】
そこで、ステップ１４で求めたピーク−ピーク値が、良品と不良品を弁別できるだけ差があるか否かを判断する（ＳＴ１５）。そして、弁別できるほど差がない場合には、ステップ１３に戻り、γの値を変更し、前後幅の再設定を行う。
【００４７】
そして、ステップ１５の分岐判断でＹｅｓとなると、ステップ１４で求めたピーク−ピーク値を特徴量して出力する（ＳＴ１６）。図９のＦＦＴ波形の微分波形に対する特徴量としては、例えば図１０に示すようになり、良品と不良品とを確実に弁別できる。
【００４８】
なお、図８に示すフローチャートでは、ピーク−ピーク値をそのまま特徴量として使用したが、図５に示したフローチャートのステップ５と同様に、適宜の変換処理をしてももちろんよい。
【００４９】
また、ステップ１５の分岐判断は、学習機能の時に実行され、調整によりステップ１３で行った前後幅が決定され、実際の検査時においては、中心周波数や前後幅を初期設定値として持ち、特徴量演算部１５における実際の処理は、ステップ１１，１４，１６を実行するようになる。なお、その他の構成並びに作用効果は、上記した第１の実施の形態と同様であるので、その詳細な説明を省略する。
【００５０】
図１１以降は、第３の実施の形態の要部を示している。上記した各実施の形態では、いずれも、波形データのピーク付近のデータを抽出し、それに基づいて各処理を行うようにしたが、本実施の形態では、それと相違して、高周波帯域を検査対象範囲としている。従って、波形抽出部１３は、良品サンプルにおける平坦な領域（図3中、Ｒ参照）を抽出する。なお、この様に抽出する周波数帯域が異なることから、フィルタ処理部１２における上限値，下限値も適宜変更するのはもちろんである。
【００５１】
そして、このように平坦部分について波形抽出された波形に対し、ＦＦＴ部１４でＦＦＴ処理すると、図１２に示すＦＦＴ波形が得られる。図１２（ａ）が良品についてのＦＦＴ波形であり、図１２（ｂ）が不良品についてのＦＦＴ波形である。
【００５２】
特徴量演算部１５は、図１１に示すフローチャートを実施する機能を有する。すなわち、まず良品サンプルにおけるＦＦＴ波形の平坦部を対象範囲に設定する（ＳＴ２１）。
【００５３】
よく知られているように、回帰直線の決定係数ｒ^２は、フラット（傾きがない）な波形では、ばらつくため、良品と不良品の判定が行いにくい。そこで、データのずらし量ｋを設定する（ＳＴ２２）。このずらし量ｋは、傾きであり、周波数が大きくなるにつれて、実際のデータに対し、徐々に大きな値を加算するようにした。つまり、本来であれば、回帰直線は、Ｘ軸に平行になるところ、ずらし量ｋの傾きを持った直線となる（図１３参照）。
【００５４】
次いで、ずらしデータに対して、回帰直線の決定係数ｒ^２を計算する（ＳＴ２３）。そして、良品のｒ^２が１に近いかを判断する（ＳＴ２４）。実際には、閾値を設け、閾値以上の場合にはＹｅｓと判断するようにする。そして、１から離れている場合には、ずらし量ｋが十分ではなかったと判断し、ステップ２２に戻り再度ずらし量の設定を行う。
【００５５】
また、良品についてのｒ^２が１に近い場合には、良品と不良品のｒ^２に差があるか否かを判断する。両者を弁別できるに十分な差がない場合には、ステップ２２に戻り、再度ずらし量を設定する。
【００５６】
そして、ステップ２４とステップ２５のいずれもが、Ｙｅｓの場合に、そのときのずらし量ｋを正規のものに設定し、求めた決定係数ｒ^２を特徴量値に変換する（ＳＴ２６）。この変換処理は、図５に示したステップ５と同様であるので、その詳細な説明を省略する。
【００５７】
そして、求めた特徴量を出力する（ＳＴ２７）。図１２のＦＦＴ波形の場合の特徴量としては、例えば図１４に示すようになり、良品と不良品とを確実に弁別できる。なお、その他の構成並びに作用効果は、上記した各実施の形態と同様であるので、その詳細な説明を省略する。
【００５８】
【発明の効果】
以上のように、この発明では、ＦＦＴ波形に対して回帰直線の決定係数などの適合度を求めたり、ＦＦＴ波形の微分値を求めることにより、ＦＦＴ波形に含まれる歪みの有無・程度を特徴量として抽出することができ、入力された波形信号に対し正確に歪みの有無・良否判定を行うことができる。
【図面の簡単な説明】
【図１】本発明の第１の実施の形態を示す図である。
【図２】本発明の第１の実施の形態を示す図である。
【図３】入力されたスイープ信号の一例を示す波形図である。
【図４】フィルタ処理後のスイープ信号を示す波形図である。
【図５】特徴量演算部の機能を示すフローチャートである。
【図６】作用を説明する図である。
【図７】抽出された特徴量値の一例を示す図である。
【図８】第２の実施の形態の要部となる特徴量演算部の機能を示すフローチャートである。
【図９】作用を説明する図である。
【図１０】抽出された特徴量値の一例を示す図である。
【図１１】第３の実施の形態の要部となる特徴量演算部の機能を示すフローチャートである。
【図１２】作用を説明する図である。
【図１３】作用を説明する図である。
【図１４】抽出された特徴量値の一例を示す図である。
【符号の説明】
１０ 検査装置
１１ Ａ／Ｄ変換部
１２ フィルタ処理部
１３ 波形抽出部
１４ ＦＦＴ部
１５ 特徴量演算部
１６ 判定部[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a waveform inspection apparatus.
[0002]
[Prior art]
In automobiles and home appliances, a rotating device incorporating a motor is very often used. For example, taking an automobile as an example, rotating equipment is mounted everywhere in the engine, power steering, power seat, mission and others. Household appliances include refrigerators, air conditioners, washing machines and various other products. When the rotating device is actually operated, a sound is generated with the rotation of the motor or the like.
[0003]
Some of these sounds are inevitably generated along with normal operations, and other sounds are generated due to defects. Examples of abnormal sounds associated with the failure include bearing abnormalities, internal abnormal contact, imbalance, and foreign matter contamination. More specifically, there are gear chipping, foreign object biting, spot flaws, and abnormal noise such that the rotating part and the fixed part inside the motor rub for a moment during rotation. Further, as sounds that people feel uncomfortable, for example, there are various sounds from 20 Hz to 20 kHz that humans can hear, for example, about 15 kHz. And when the sound of the predetermined frequency component is generated, it becomes an abnormal sound. Of course, abnormal sounds are not limited to this frequency.
[0004]
The sound associated with such a defect is not only unpleasant, but may cause further failure. Therefore, for the purpose of quality assurance for each of these products, production factories usually perform “sensory inspection” that relies on the five senses such as hearing and tactile sensation, and determine the presence or absence of abnormal sounds. Specifically, it is done by listening with the ear or touching it with the hand to check the vibration. The sensory test is defined by the sensory test term JIS Z8144.
[0005]
By the way, in the sensory test that relies on the five senses of such an inspector, not only skill is required, but also the determination results vary widely such as individual differences and changes with time. Furthermore, there is a problem that it is difficult to convert the determination results into data and numerical values and to manage them. Therefore, in order to solve such a problem, there is an abnormal sound inspection apparatus for the purpose of stable inspection based on a quantitative and clear standard. This abnormal sound inspection system is a device for the purpose of automating the “sensory inspection” process, and measures vibration and sound of the product drive unit with a sensor, and uses a frequency analysis device that applies the analog signal to the analog signal. Thus, the frequency component is examined and inspected (Patent Documents 1, 2, and 3). The analog signal may be analyzed by applying a band pass filter.
[0006]
The technology disclosed in Patent Documents 1 to 3 will be briefly described. A frequency analysis apparatus to which the FFT algorithm is applied can analyze a frequency domain of a time domain signal by a fast Fourier transform algorithm. On the other hand, the frequency range of abnormal sounds is also determined to some extent. Accordingly, since the component corresponding to the abnormal sound generation region can be extracted from the frequency components extracted by the analysis, the feature amount of the extracted component is obtained. Then, the presence / absence of the abnormality and the cause thereof are estimated from the feature amount using fuzzy inference.
[0007]
In the abnormal sound inspection system described above, automatic determination can be made in accordance with a predetermined standard, and the inspection result (actual result) and waveform data at that time can be stored in a storage device in the abnormal sound inspection system. .
[0008]
[Patent Document 1]
JP-A-11-173756
[Patent Document 2]
JP-A-11-173909
[Patent Document 3]
JP 2001-91414 A
[0009]
[Problems to be solved by the invention]
In the above-described conventional apparatus, it is possible to inspect abnormal noises emitted from rotating devices such as motors and engines relatively correctly. However, when the above abnormal sound inspection system is applied to a device that applies a sweep signal to a speaker and determines whether or not the sound output from the speaker is distorted, for example, as in a quality inspection of a speaker, an accurate determination Could not do.
[0010]
In other words, the sweep sound used for speaker quality inspection or the like is obtained by gradually changing the frequency from a low frequency to a high frequency, and the sound will be detected by a microphone and FFT conversion will be described in the embodiment. It becomes a unique shape waveform continuously. Therefore, it has been impossible to judge whether or not the feature amount of the system is based on the assumption that the spectrum of the conventional predetermined frequency appears discontinuously. For this reason, in speaker quality inspections and the like, inspections still relying on the five senses of the inspector are still mainstream, and there are large variations in judgment results such as individual differences and changes with time.
[0011]
It is an object of the present invention to provide a waveform inspection apparatus that can accurately determine the presence / absence / non-defectiveness of distortion even if the inspection target is a sweep signal.
[0012]
[Means for Solving the Problems]
The waveform inspection apparatus according to the present invention obtains a regression line for an FFT processing means for performing FFT processing on a waveform signal based on an input sweep waveform signal, and a predetermined frequency range of the FFT waveform generated by the FFT processing. And a feature amount calculation means for obtaining information based on the degree of fitness for the regression line as a feature amount, and a determination means for judging normality / abnormality based on the obtained feature amount. In the embodiment, the fitness is a coefficient of determination r. ² However, any coefficient other than the determination coefficient may be used as long as the degree of deviation and variation of each data constituting the FFT waveform from the regression line can be obtained.
[0013]
Further, as another solution means, an FFT processing means for performing FFT processing on the waveform signal based on the input sweep waveform signal, and obtaining a differential value of each frequency of the FFT waveform generated by the FFT processing, A feature amount calculation unit that obtains the degree of variation of the obtained differential value as a feature amount and a determination unit that determines normality or abnormality based on the obtained feature amount may be provided. The variation of the differential value is obtained from the peak-peak value in the embodiment, but the present invention is not limited to this, and various other dispersions can be used.
[0014]
And in each above-mentioned invention, FFT processing can be performed with respect to the part of a fixed period from the position shifted predetermined amount from the peak position of the said sweep waveform signal. Note that shifting by a predetermined amount includes both positive and negative directions.
[0015]
As another embodiment, FFT processing means for performing FFT processing on a flat region of the waveform signal based on the input sweep waveform signal, and a predetermined frequency range of the FFT waveform generated by the FFT processing On the other hand, a feature amount calculation means that obtains a regression line from a corrected FFT waveform obtained by adding a correction value that increases with increasing frequency, and obtains information based on the degree of fitness for the regression line as a feature amount And determination means for determining normality / abnormality based on the obtained feature amount. In the embodiment, the correction value corresponds to the shift amount k. In the embodiment, k is positive and is gradually increased. However, a negative value may be used.
[0016]
In the case of the FFT waveform for the sweep signal, since there is a value at each frequency, the waveform is continuous. When normal, it changes gently and the amount of change becomes almost constant. However, when distortion occurs, the amount of change increases at the distortion part.
[0017]
Therefore, when the regression line is obtained for the FFT waveform, each data exists in the vicinity of the regression line when it is normal, but when it is defective / abnormal, it greatly deviates from the regression line at the distortion portion. Accordingly, it is possible to determine whether the regression line is good or bad by determining the goodness of fit (decision coefficient). This is the first invention, which is realized by the first embodiment.
[0018]
However, when the slope of the regression line becomes small, the degree of fitness varies even if it is normal, so correctness can be determined by forcibly tilting the regression line after correction. This is the third invention, which is realized by the third embodiment.
[0019]
Further, as described above, since the change of the FFT waveform is large in the portion where the distortion is generated, the derivative of the FFT waveform is obtained, and the larger the variation of the differential value, the more the distortion occurs and it can be said that the defect is abnormal. .
[0020]
DETAILED DESCRIPTION OF THE INVENTION
1 and 2 show a first embodiment of the present invention. As shown in FIG. 1, an output of a sweep signal generator 1 that generates a sweep signal that is gradually and gradually changed from a low frequency to a high frequency is input to a speaker 2. The sweep signal is preferably a sin waveform, and a waveform signal in which the wavelength continuously changes with a constant amplitude is good. As a result, a sweep sound whose frequency gradually increases (becomes high) from the low frequency (low sound) is output from the speaker 2.
[0021]
The sound generated from the speaker 2 is collected by the microphone 3 and the signal level is amplified by the amplifier 4 and then input to the inspection device 10. This inspection apparatus 10 determines whether or not the sweep sound is distorted.
[0022]
As shown in FIG. 2, the inspection apparatus 10 includes an A / D conversion unit 11 that converts an analog sweep signal amplified through an amplifier 4 into a digital signal, and a signal output from the A / D conversion unit 11. A filter processing unit 12 that removes unnecessary components such as noise components, a waveform extraction unit 13 that extracts a waveform of a predetermined period (1 frame: about 0.1 second) from the filtered waveform, and a waveform extraction unit 13 The FFT unit 14 that performs frequency domain analysis of the time domain signal using the fast Fourier transform algorithm, and the feature amount calculation unit that calculates a predetermined feature amount from the FFT waveform generated by the FFT unit 14. 15 and a determination unit 16 that performs pass / fail determination based on the feature amount extracted by the feature amount calculation unit 15. The inspection apparatus 10 can be realized by incorporating the above various functions into a personal computer, for example. And if there is what is already mounted in the personal computer, the A / D conversion part 11 can utilize it. And it can comprise by installing the software program (application) provided with the other function in a personal computer. Of course, not all of them can be realized by software, but a part of them can also be realized by hardware.
[0023]
Next, the function of each part will be described. However, the internal processing algorithms in the A / D conversion unit 11 and the FFT unit 14 are conventionally known, and there is no parameter adjustment, and therefore detailed description thereof is omitted.
[0024]
First, the filter processing unit 12 is configured by a band pass filter in the present embodiment, and passes a frequency range of, for example, 500 to 1500 Hz. Thereby, for example, an input waveform as shown in FIG. 3 is converted into a waveform as shown in FIG. Note that the upper limit value and the lower limit value of the passband, which are parameters of the filter processing unit 12, can be changed via an input device (not shown) provided in the inspection apparatus 10.
[0025]
The graphs shown in FIGS. 3 and 4 will be described. In both graphs, the vertical axis represents the voltage of the input waveform (that is, the amplitude of the waveform signal, in other words, the strength of the signal), and the horizontal axis represents time. . Referring to FIG. 3, for example, a signal of 10 Hz is given to the speaker from the time around t = 3000, and then the frequency is gradually increased to 500 Hz near t = 4000, 1500 Hz near t = 6500, t = 15000. Assume that a 50 kHz signal is given in the vicinity. As a result, as shown in FIG. 3, it can be seen that sound starts to be heard from the speaker at around t = 4000 and no sound is produced at around t = 11000. FIG. 3 can be said to represent the sound reproduction characteristics of the speaker.
[0026]
The waveform extraction unit 13 has a function of extracting a partial waveform to be analyzed from the entire waveform after passing through the filter. This is for extracting only an appropriate part so that it is easy to inspect whether the speaker to be inspected is a good product or a defective product. In FIG. 4, the portion illustrated by the width of t is extracted to be an inspection target. The area to be extracted is determined in advance as a region including a portion where the difference is generated by previously analyzing a portion where the difference between the non-defective product and the defective product is generated. The example here will be described with reference to FIG. 6, but since it is known in advance analysis that there is a possibility that the speaker emits abnormal noise at 800 Hz to 900 Hz, it is good to inspect the surrounding characteristics. It can be said that it is suitable for defect determination. Therefore, a waveform including a period during which the speaker may reproduce abnormal noise is extracted. Then, it is known from the characteristics that if it is attempted to include a period in which the speaker reproduces the sound at 800 Hz to 900 Hz, it may be taken in a little later from the peak value P of the characteristic graph. Therefore, since it is known in advance that the waveform of the period t should be extracted a little after the peak value P shown in FIG. 4 is issued, the region extracted by the waveform extraction unit so as to capture the waveform of that portion. Are defined as “α” as the start position and “t” as the period. Actually, in FIG. 4, the start position α is set to time 4600, and t is set to 300 (actual value is about 0.1 seconds). For example, when the frequency at which distortion is likely to occur (frequency band in which abnormal noise may occur) is known, this setting is set to a central position α such that the frequency is at the center of the frame, and the previous period t1. , The later period t2 (that is, the period before and after) may be taken in. Also, if the distortion occurrence position is not known in advance, if there is non-defective / defective sample data, etc., determination processing is performed while changing α based on the sample data, and the accuracy of the determination result reaches the standard. It can be determined by adjusting as follows.
[0027]
The FFT unit 14 performs a Fourier transform process on the waveform extracted by the waveform extraction unit 13. That is, the waveform of the portion of the period t in FIG. 4 is Fourier transformed to be as shown in FIG. The horizontal axis of FIG. 6 is the frequency, which indicates the frequency of the sound that the speaker is reproducing. The vertical axis represents the amplitude of the reproduced sound of the speaker and the sound intensity.
[0028]
The feature amount calculation unit 15 has a function of executing the flowchart shown in FIG. This flowchart shows a learning function at the time of parameter setting and a function to be performed at the time of actual defect determination (inspection).
[0029]
First, the center frequency for the acquired FFT waveform is set (ST1). That is, FIG. 6 shows an example of the FFT waveform. Here, FIG. 6A shows a non-defective FFT waveform, and FIG. 6B shows a defective FFT waveform. Note that the vertical axis in FIG. 6 is the spectrum power (that is, the intensity of the reproduced sound of the speaker, in other words, the waveform amplitude), and the horizontal axis is the frequency.
[0030]
As is clear from FIG. 6, each FFT waveform has a continuous mountain-shaped waveform that rises once and then descends as the frequency increases, and the non-defective FFT waveform has almost a certain interval when descending. In contrast to the linear shape, the FFT waveform of a defective product is distorted (indented in the example in the figure) in the above-mentioned fixed interval. More specifically, it can be seen that the sound reproduction characteristics of 10 Hz (around 810 Hz to 830 Hz) around the frequency around 820 Hz are different from the non-defective products. It should be noted that the vertical axis of the characteristic graph shown here is the amplitude (reproduction sound intensity), which is depressed on the graph as compared with the non-defective product, but sounds abnormally or uncomfortable to the human ear.
[0031]
Therefore, a frequency that is in the middle of the above-described fixed interval is set as the center frequency. Specifically, for example, the FFT waveform shown in FIG. 6 may be output to the display device, and the user may determine the center frequency and input it numerically, or a guide bar extending in the vertical axis direction as shown in the figure. The center frequency can also be set by displaying them in a superimposed manner, operating the input device to move the guide bar to the left and right, and determining it at a predetermined position. Furthermore, when the frequency at which distortion occurs is known in advance, the value can be set as the center frequency. In the present embodiment, the center frequency is set to 900 Hz.
[0032]
Next, the number of data used for the calculation is determined (ST2). In the present embodiment, since the center frequency f is equal to the front and rear, “β” is substituted via the input device, and “f−β” and “f + β” are calculated in response thereto. Determine the front and rear width. Β input at this time is an adjustment parameter.
[0033]
For the data in the range of “f−β” to “f + β”, the horizontal axis (frequency) X: horizontal axis (frequency) and Y: vertical axis (power spectrum) are substituted into the following formula, Coefficient of determination r ² Is calculated (ST3).
[0034]
[Expression 1]

[0035]
Coefficient of determination r ² Indicates the fitness of the regression line and takes a value between 0 and 1. And r ² = 1 means perfect fit. Accordingly, as in the case of a non-defective FFT waveform shown in FIG. 6A, when the power spectrum decreases almost linearly as the frequency increases, r ² Takes a value close to 1, and if there is a distorted part like the FFT waveform of the defective product shown in FIG. ² Becomes a value away from 1.
[0036]
Therefore, r of good product and defective product obtained in step 3 ² However, it is determined whether or not there is a difference as much as possible to distinguish the two (ST4). If there is no difference that can be discriminated, the process returns to step 2 to change the value of β and reset the front-rear width.
[0037]
Then, when the branch determination at step 4 is Yes, the obtained determination coefficient r ² Is converted into a feature value (ST5). That is, as described above, the non-defective product has a larger value closer to 1 in the present embodiment. However, in the present embodiment, the feature value in the case of a non-defective product due to the relationship with other feature values (the description is omitted). Therefore, the coefficient of determination is converted to a feature value by the following conversion formula to match it.
Feature value = 1000− (r ² × 1000)
Then, the obtained feature amount is output (ST6). The feature quantity in the case of the FFT waveform in FIG. 6 is, for example, as shown in FIG.
[0038]
Of the processing steps in the flowchart shown in FIG. 5, the conversion in step 5 is not limited to the above-described one. ² In addition to multiplying by a predetermined constant, various conversions can be performed. Further, when the determination coefficient is used as it is, the processing in step 5 itself may be omitted.
[0039]
Further, the branch determination in step 4 is executed at the time of the learning function, and the front / rear width performed in step 2 is determined by adjustment. In actual inspection, the center frequency and the front / rear width are set as initial setting values, and the feature amount is determined. In the actual processing in the calculation unit 15, steps 3, 5, and 6 are executed.
[0040]
The determination unit 16 performs determination processing based on the feature amount output from the feature amount calculation unit 15, and can perform pass / fail determination by threshold processing, for example. Of course, when a plurality of types of feature quantities are output, they can be comprehensively determined.
[0041]
The processing algorithm of the feature amount calculation unit 15 is not limited to that of the first embodiment described above, and various modifications can be made. For example, it can be set as the function which implements the flowchart shown in FIG. 8 which is 2nd Embodiment. This flowchart also includes a learning function.
[0042]
First, the acquired FFT waveform is differentiated (ST11). That is, as described above, since the FFT waveform in the case of a non-defective product is substantially linear, variation in the deformation amount is small. On the other hand, since the FFT waveform in the case of a defective product has distortion and depression, variation in the amount of change becomes large. Therefore, the feature amount is extracted based on the differential value. By executing such processing, for example, the FFT waveforms shown in FIGS. 6A and 6B become as shown in FIGS. 9A and 9B, respectively. In FIG. 9, the horizontal axis represents the frequency, and the vertical axis represents the differential value of the spectrum power of the sound reproduction of the speaker.
[0043]
Next, the center frequency for the differentiated waveform of the acquired FFT waveform is set (ST12). This center frequency is also set as a parameter by operating the input device, as in the above-described embodiment. That is, if the frequency at which distortion occurs is known in advance, the value can be set as the center frequency. In the present embodiment, the center frequency is set to 900 Hz. Further, in the present embodiment, in the case of a defective product, distortion occurs in the section where the FFT waveform is falling. For example, the differential waveform shown in FIG. As a center frequency, an appropriate position in a section where the differential value is negative is determined. Then, the determined value may be input as a numerical value, or as shown in the figure, a guide bar extending in the vertical axis direction is displayed in an overlapping manner, and the input bar is operated to move the guide bar to the left and right, and a predetermined position. The center frequency can also be set by determining in step (1).
[0044]
Next, the number of data used for calculation is determined (ST13). In the present embodiment, the center frequency f1 is equal to the front and rear, so that “γ” is substituted via the input device, and “f1−γ” and “f1 + γ” are calculated by receiving this. Determine the front and rear width. Γ input at this time is an adjustment parameter.
[0045]
Then, the peak-peak value in the target range of “f1−γ” to “f1 + γ” is obtained (ST14). That is, as described above, the variation in the differential value of the FFT waveform in the case of a non-defective product is small, but the differential value of the FFT waveform in the case of a defective product is positive or negative at the distortion / dent portion, and the variation in the variation is large. Accordingly, the peak-peak value is smaller for non-defective products and larger for defective products.
[0046]
Therefore, it is determined whether or not the peak-peak value obtained in step 14 is different enough to discriminate between a non-defective product and a defective product (ST15). If there is no difference that can be discriminated, the process returns to step 13 to change the value of γ and reset the front and rear width.
[0047]
When the branch determination in step 15 is Yes, the peak-peak value obtained in step 14 is output as a feature amount (ST16). For example, as shown in FIG. 10, the feature amount with respect to the differential waveform of the FFT waveform in FIG. 9 can be reliably distinguished from a non-defective product.
[0048]
In the flowchart shown in FIG. 8, the peak-to-peak value is used as it is as the feature quantity, but an appropriate conversion process may be performed as in step 5 of the flowchart shown in FIG.
[0049]
Further, the branch determination in step 15 is executed at the time of the learning function, and the front and rear width made in step 13 is determined by adjustment. At the actual inspection, the center frequency and the front and rear width are set as initial setting values, and the feature amount In the actual processing in the calculation unit 15, steps 11, 14, and 16 are executed. Since other configurations and operational effects are the same as those in the first embodiment, detailed description thereof is omitted.
[0050]
FIG. 11 and subsequent figures show the main part of the third embodiment. In each of the above-described embodiments, data near the peak of the waveform data is extracted and each processing is performed based on the extracted data. However, in this embodiment, unlike the above, the high frequency band is inspected. The range. Therefore, the waveform extraction unit 13 extracts a flat region (see R in FIG. 3) in the non-defective sample. Since the frequency bands to be extracted are different in this way, the upper limit value and the lower limit value in the filter processing unit 12 are of course changed as appropriate.
[0051]
Then, when the FFT processing is performed on the waveform extracted in this way for the flat portion by the FFT unit 14, an FFT waveform shown in FIG. 12 is obtained. FIG. 12A shows an FFT waveform for a non-defective product, and FIG. 12B shows an FFT waveform for a defective product.
[0052]
The feature amount calculation unit 15 has a function of executing the flowchart shown in FIG. That is, first, the flat part of the FFT waveform in the non-defective sample is set as the target range (ST21).
[0053]
As is well known, the coefficient of determination r of the regression line ² However, it is difficult to determine whether the product is non-defective or defective because the waveform varies with a flat waveform (having no inclination). Therefore, a data shift amount k is set (ST22). This shift amount k is an inclination, and as the frequency increases, a larger value is gradually added to the actual data. That is, originally, the regression line is a straight line having an inclination of the shift amount k in parallel with the X axis (see FIG. 13).
[0054]
Next, a regression line determination coefficient r is applied to the shift data. ² Is calculated (ST23). And good r ² Is determined to be close to 1 (ST24). Actually, a threshold value is provided, and if it is equal to or greater than the threshold value, Yes is determined. If it is away from 1, it is determined that the shift amount k is not sufficient, and the process returns to step 22 to set the shift amount again.
[0055]
Also, r for good products ² When r is close to 1, r for good and defective products ² To determine whether there is a difference between If there is not a sufficient difference to discriminate the two, the process returns to step 22 and the shift amount is set again.
[0056]
If both Step 24 and Step 25 are Yes, the shift amount k at that time is set to a normal one, and the obtained determination coefficient r ² Is converted into a feature value (ST26). Since this conversion process is the same as step 5 shown in FIG. 5, detailed description thereof is omitted.
[0057]
Then, the obtained feature amount is output (ST27). The feature quantity in the case of the FFT waveform of FIG. 12 is as shown in FIG. 14, for example, and it is possible to reliably discriminate between good products and defective products. Since other configurations and operational effects are the same as those of the above-described embodiments, detailed description thereof is omitted.
[0058]
【The invention's effect】
As described above, according to the present invention, the presence / absence / degree of distortion included in the FFT waveform is obtained by obtaining the degree of conformity such as the coefficient of determination of the regression line with respect to the FFT waveform or by obtaining the differential value of the FFT waveform. And the presence / absence / non-defectiveness of the distortion can be accurately determined for the input waveform signal.
[Brief description of the drawings]
FIG. 1 is a diagram showing a first embodiment of the present invention.
FIG. 2 is a diagram showing a first embodiment of the present invention.
FIG. 3 is a waveform diagram showing an example of an input sweep signal.
FIG. 4 is a waveform diagram showing a sweep signal after filter processing.
FIG. 5 is a flowchart illustrating functions of a feature amount calculation unit.
FIG. 6 is a diagram illustrating an operation.
FIG. 7 is a diagram illustrating an example of extracted feature value.
FIG. 8 is a flowchart illustrating functions of a feature amount calculation unit that is a main part of the second embodiment;
FIG. 9 is a diagram illustrating an operation.
FIG. 10 is a diagram illustrating an example of extracted feature value.
FIG. 11 is a flowchart illustrating a function of a feature amount calculation unit that is a main part of the third embodiment;
FIG. 12 is a diagram illustrating an operation.
FIG. 13 is a diagram illustrating an operation.
FIG. 14 is a diagram illustrating an example of extracted feature value.
[Explanation of symbols]
10 Inspection equipment
11 A / D converter
12 Filter processing section
13 Waveform extraction unit
14 FFT section
15 Feature value calculator
16 Judgment part

Claims (4)

FFT processing means for performing FFT processing on the waveform signal based on the input sweep waveform signal;
A feature amount computing means for obtaining a regression line with respect to a predetermined frequency range of the FFT waveform generated by the FFT processing, and obtaining information based on the degree of fitness for the regression line as a feature amount;
A waveform inspection apparatus comprising a determination means for determining normality or abnormality based on the obtained feature amount. FFT processing means for performing FFT processing on the waveform signal based on the input sweep waveform signal;
A feature amount computing means for obtaining a differential value of each frequency of the FFT waveform generated by the FFT processing and obtaining a degree of variation of the obtained differential value as a feature amount;
A waveform inspection apparatus comprising a determination means for determining normality or abnormality based on the obtained feature amount. The waveform inspection apparatus according to claim 1, wherein the FFT processing is performed on a portion of a certain period from a position shifted by a predetermined amount from a peak position of the sweep waveform signal. FFT processing means for performing FFT processing on a flat region of the waveform signal based on the input sweep waveform signal;
A regression line is obtained for a corrected FFT waveform obtained by adding a correction value that increases with increasing frequency to a predetermined frequency range of the FFT waveform generated by the FFT processing, and conformity to the regression line is obtained. A feature amount calculating means for obtaining information based on the degree as a feature amount;
A waveform inspection apparatus comprising a determination means for determining normality or abnormality based on the obtained feature amount.

Images