A method for identifying the grinding state of a food processor based on vibration signal analysis

Abstract

This invention discloses a method for identifying the grinding state of a food processor based on vibration signal analysis, belonging to the field of automated measurement technology. The method includes: First, collecting vibration signals from the food processor under different operating conditions using an accelerometer and performing data preprocessing. Second, performing FFT and STFT transformations on the signals to analyze frequency domain characteristics. Sliding window sampling is then used to calculate sixteen feature indicators for each sample. Finally, based on these indicators, a model is used to accurately classify the training set for different operating conditions. The identification is verified using test set samples, thus achieving accurate identification of the food's state during grinding. This invention can accurately determine whether the food being blended has reached a steady state, while preventing the machine from continuously grinding without food and allowing for reasonable control of start-stop times. These findings will provide important references and guidance for the further optimization design of blending household appliances such as food processors, soy milk makers, and high-speed blenders.

Description

Technical Field

[0001] This invention belongs to the field of automated measurement technology, and in particular relates to a method for identifying the grinding state of a food processor based on vibration signal analysis. Background Technology

[0002] With the continuous upgrading of public health awareness and consumption concepts, the nutrition and health industry in various regions has entered a period of rapid development. Food processors, simple to operate and highly practical, integrate functions such as making soy milk, grinding dry powders, juicing, mincing meat, and shaved ice, making them one of the essential kitchen appliances and playing an increasingly important role in modern life. Food processors utilize the high-speed rotation of blades in the blending cup to continuously mix, grind, and chop ingredients to achieve the desired food texture. However, in actual use, users encounter numerous problems. In recent years, many researchers have conducted extensive research and optimization on the issue of excessive vibration and noise during the grinding process of food processors.

[0003] However, besides the issue of excessive machine noise, another key problem is the inability to judge the grinding state of the target material in real time. The machine's operating time relies entirely on the user's subjective judgment, easily leading to insufficient or excessive grinding, affecting the taste and quality of food, and causing unnecessary energy waste. Especially during blending, solids such as fruits and ice cubes can swirl, causing the machine to run dry. While some blenders allow setting fixed start and stop times, this makes it impossible to properly control the blending effect. Furthermore, different types and hardnesses of food have varying degrees of difficulty in crushing, and combined with different blending speeds, if the user fails to accurately assess the hardness and quality of the food each time, inaccurate or untimely judgments of the degree of food crushing will occur, ultimately affecting the edible experience. High-speed rotating blades also generate heat, which can spoil food if blending is prolonged. Therefore, accurately identifying the state of the target material and mastering the start and stop times of the blender is extremely important for different types of food. Thus, in-depth research into the vibration signal characteristics of blenders during operation and the identification of the target material's state to optimize blender efficiency is of significant research importance. Summary of the Invention

[0004] The purpose of this invention is to provide a method for identifying the grinding state of a food processor based on vibration signal analysis, so as to solve the technical problem of being unable to determine the grinding state of a target material in real time.

[0005] To achieve the above objectives, the specific technical solution of the food processor grinding state identification method based on vibration signal analysis of the present invention is as follows:

[0006] A method for identifying the grinding state of a food processor based on vibration signal analysis includes the following steps, which are performed sequentially:

[0007] (a) A miniature accelerometer was used to collect the raw vibration signals of the blender under different operating conditions. Different operating conditions refer to different ingredients being ground inside the blender. The experiment was conducted by installing miniature accelerometers at four different locations: on the circuit board of the blender base, near the base, slightly above the base, and outside the base. Vibration signal data were collected from these locations. The installation positions of the miniature accelerometers were compared to ① the environmental noise level (the sensor itself introduces noise, which reduces data quality. Noise interferes with the required signal, so a lower noise level is generally considered to be better data quality) and ② the degree of physical contact (a better location should provide stable physical contact to maximize the transmission of vibration signals). Alternatively, the time-domain plot of the data can be directly viewed; the larger the overall vibration amplitude and the more obvious the change, the higher the data quality. The miniature accelerometer was used to collect vibration signal data of the blender under different operating conditions.

[0008] (b) Vibration data was selected from the data of the three axes of the miniature accelerometer (vibrations on different axes represent motion in different directions. The X-axis usually represents left-right motion, the Y-axis represents up-down motion, and the Z-axis represents forward-backward motion. Obviously, the operation of the food processor is closest to left-right vibration, and the vibration amplitude of the X-axis is larger than that of the three axes, which may contain more information). The vibration data of the X-axis direction of the sensor was of the highest quality and most representative in the experiment. The length of 60,000 valid data points was selected for each working condition within the effective time period to facilitate further processing and analysis. The raw vibration signal data collected in step (a) was processed by sliding window to select the effective time period of the vibration signal (due to limitations of signal acquisition conditions and equipment, such as data collected before the machine is turned on, the raw vibration data collected contains a certain period of invalid components, that is, the segment where the vibration data is close to 0, which is the segment where the signal is missing or incomplete. Therefore, the effective time period is the period in which the amplitude of the vibration data changes continuously within a certain range).

[0009] (c) For the non-stationary signal generated by the vibration during the operation of the food processor, the characteristics of the time domain and frequency domain of a small segment (five cycles) of vibration signal in the effective time period under different working conditions are compared and analyzed by Fast Fourier Transform (FFT), as well as the different frequency components and the different distribution of signal energy in the frequency domain.

[0010] (d) Based on step (c), continue to perform a short-time Fourier transform (STFT) on a small segment (five cycles) of vibration signal in the effective time period under different working conditions to obtain information in the time and frequency domain.

[0011] (e) The vibration signal data of the effective time period under different working conditions obtained in step (b) are segmented, and sliding window sampling is performed with data of fixed length. The ratio of training set samples to test set samples is 5:2 for each working condition. Then, the data is evenly segmented with the four harmonics of the frequency domain information obtained in step (c) as the center, and the maximum frequency value and average amplitude in the four frequency bands of each sample data segment are calculated respectively.

[0012] Step (f) calculates eight feature functions for all sample data under different operating conditions obtained by sliding window sampling in step (e). The eight feature functions include four typical time-domain feature parameters and four typical frequency-domain feature parameters. The four typical time-domain feature parameters are: variance, peak factor, impulse factor, and kurtosis. The four typical frequency-domain feature parameters are: mean spectral amplitude, spectral energy, centroid frequency, and frequency variance. There are a total of eight typical time-domain and frequency-domain feature functions. Step (e) has already calculated the maximum frequency value and average amplitude in four frequency bands for each sample, totaling eight feature values. This step calculates the eight typical time-domain and frequency-domain feature function values ​​for each sample, which is equivalent to calculating 16 feature values ​​for each sample data obtained by sliding window sampling.

[0013] Step (g) uses 16 feature indicators—the maximum frequency, the average amplitude, four typical time-domain feature parameters, and four typical frequency-domain feature parameters—as training features. These are then sorted according to the different working conditions represented by the samples, i.e., the hardness level of the food, and numerically labeled accordingly. All samples in both the training and test sets are then labeled using a Support Vector Machine (SVM) model. This model is then used to train all the training set samples from step (e), linking the model to the labels and obtaining a classification result for different working conditions.

[0014] Steps (h) and (g) are based on the classification results trained in step (g). Using all samples from the test set in step (e), each sample is processed by a support vector machine model for identification, and the accuracy of each sample's identification is observed. If a 100% accuracy rate is achieved, the model can identify a segment of vibration signal data to obtain a label number, thereby accurately determining which food's state corresponds to the current grinding state of the food processor. Specifically, the hardness of different working conditions is distinguished based on the differences between the maximum frequency and average amplitude values ​​within the four frequency bands calculated in step (e) and the 16 feature values ​​of the eight feature functions calculated in step (f). Five cross-validations are performed to enable the model to better classify the training samples. This minimizes classification errors in the training samples (e.g., working condition 1 associated with label 1, working condition 2 associated with label 2, etc.), and the final classification accuracy reflects the degree of association between samples and labels under this model. Then, each sample in the test set is used to verify that a 100% accuracy rate can be obtained. For example, inputting a sample data from working condition 3 in the test set can output the label number "3". Therefore, a model can be used to identify a segment of vibration signal data to obtain a tag number, thereby accurately determining which ingredient's state corresponds to the current grinding state of the food processor.

[0015] The present invention provides a method for identifying the grinding state of a food processor based on vibration signal analysis, which has the following advantages: Directly collected vibration signal data under different operating conditions is accurate and reliable; analysis of features within each domain of the vibration signal and extraction and classification of all features result in high accuracy. This method, based on vibration signal feature analysis of the food processor's grinding state, is not only accurate, reliable, and effective, but it can also effectively solve the problem of insufficient or excessive grinding time, especially for ingredients of different hardness and quality, allowing for reasonable setting of start and stop times based on the real-time grinding state of the ingredients. Furthermore, the improvement cost is low, providing significant reference value for further optimization of food processor design. Attached Figure Description

[0016] Figure 1 This is an experimental setup (72-ounce household blender) in Embodiment 1 of the present invention, which is a method for identifying the grinding state of a blender based on vibration signal analysis.

[0017] Figure 2 This is a miniature accelerometer sensor used in Embodiment 1 of the present invention, which is a method for identifying the grinding state of a food processor based on vibration signal analysis.

[0018] Figure 3 This is a comparison of signals under different operating conditions before and after FFT in Embodiment 1 of the present invention, which is a method for identifying the grinding state of a food processor based on vibration signal analysis.

[0019] Figure 4 This is a comparison of signal time-frequency diagrams under three working conditions in Embodiment 1 of the present invention, which is a method for identifying the grinding state of a food processor based on vibration signal analysis.

[0020] Figure 5 This is the confusion matrix result of SVM training under three working conditions in Embodiment 1 of the present invention, which is a method for identifying the grinding state of a food processor based on vibration signal analysis.

[0021] Figure 6 This is the confusion matrix result of SVM training under nine working conditions in Embodiment 1 of the present invention, which is a method for identifying the grinding state of a food processor based on vibration signal analysis.

[0022] Figure 7 This is an overall flowchart of a method for identifying the grinding state of a food processor based on vibration signal analysis, according to the present invention.

[0023] Figure 8 This is a box plot of the maximum frequency values ​​of each frequency band under different operating conditions in Embodiment 1 of the present invention, which is a method for identifying the grinding state of a food processor based on vibration signal analysis.

[0024] Figure 9 This is a box plot of the average amplitude of each frequency band under different operating conditions in Embodiment 1 of the present invention, which is a method for identifying the grinding state of a food processor based on vibration signal analysis.

[0025] Figure 10 This is a box plot of the signal time-domain characteristic function under different working conditions in Embodiment 1 of the present invention, which is a method for identifying the grinding state of a food processor based on vibration signal analysis.

[0026] Figure 11 This is a box plot of the frequency domain characteristic function of the signal under different working conditions in Embodiment 1 of the present invention, which is a method for identifying the grinding state of a food processor based on vibration signal analysis.

[0027] Table 1 is a detailed classification of nine different working conditions according to state type in Embodiment 1 of the food processor grinding state identification method based on vibration signal analysis according to the present invention. Detailed Implementation

[0028] To better understand the purpose, structure, and function of this invention, the following detailed description of a food processor grinding state identification method based on vibration signal analysis, in conjunction with the accompanying drawings, is provided.

[0029] This invention aims to accurately distinguish the type and quality of ingredients in a food processor, accurately determine the state of the target material being blended, monitor in real time whether the blended material has reached a steady state, and intelligently display the degree of food breakage within the food processor. This can be further developed into a real-time embedded monitoring system, and this user-friendly design can be applied to blending household appliances such as soy milk makers, juicers, and grinders. This will make the operation more convenient for users and is of great significance in improving people's quality of life.

[0030] Example 1:

[0031] See Figure 7 This embodiment discloses a method for identifying the grinding state of a food processor based on vibration signal analysis. The specific implementation includes the following steps:

[0032] Referring to step (a), based on a 72-ounce home blender, as follows: Figure 1 As shown. The data acquisition device is a high-precision miniature accelerometer, as follows. Figure 2 As shown, through testing, accelerometers were first installed at four different locations on the blender: on the base circuit board, near the base, slightly above the base, and on the outside of the base, to collect vibration signal data. Comparison revealed that when the sensor was installed near the base, the ambient noise level was lower, the physical contact was greater, the overall vibration amplitude was larger, the vibration signal data quality was higher, and the classification and recognition effect was excellent. Therefore, the optimal location for data collection was ultimately chosen: near the blender base. The experiment measured the vibration signal data of the blender under different operating conditions.

[0033] Referring to step (b), vibration signals from the blender were collected using an accelerometer under three operating conditions: an empty blender, a full blender, and a blender with ice. Since the X-axis typically represents left-right motion, the Y-axis represents up-down motion, and the Z-axis represents forward-backward motion, the blender's operation most closely resembles left-right vibration. Furthermore, the X-axis vibration data has the largest amplitude and contains more vibration information among the three axes. The X-axis vibration data, representing the highest quality and most representative data, was then selected from the data across the three axes. Due to limitations in signal acquisition conditions and equipment, the raw vibration data contained invalid components for a certain period. Therefore, a sliding window processing was performed on the raw vibration signals to select the effective periods—those where the vibration amplitude remained within a certain range and continuously varied. A length of 60,000 effective data points was uniformly selected for each operating condition to facilitate further processing and analysis.

[0034] Referring to step (c), regarding the non-stationary signal generated by the vibration during the operation of the food processor, firstly, extract a small segment of the original vibration signal under different operating conditions, specifically taking the length of 5 cycles (frequency 1344Hz, a total of 6720 data points), and observe their respective time-domain information. See [link / reference]. Figure 3 On the left, it can be observed that the food processor vibrates most strongly when ice is present, and maintains a large amplitude, with the peak value consistently around 20 m / s. 2 The signal amplitude is minimal when the glass is full of water, with the peak-to-peak value consistently around 10 m / s. 2 The amplitude is moderate when the cup is empty, with the peak-to-peak value averaging around 15 m / s. 2 Left and right. A simple way to distinguish them is by utilizing the different amplitudes of energy generated by the vibrations of the food processor under different operating conditions.

[0035] Then, a small segment of the vibration signal from step (c) is subjected to Fast Fourier Transform (FFT) to preliminarily analyze the variation patterns of the signal in the time and frequency domains under the three operating conditions. The transform is defined as follows:

[0036]

[0037] Spectral analysis can reveal information not visible in the time domain. (See also...) Figure 3 The frequency domain plot on the right reveals the different frequency components of the vibration signal under the three operating conditions, as well as the different distributions of signal energy in the frequency domain. It is evident that the frequency ranges in which several main frequency components appear under the three operating conditions are concentrated in four areas, and the magnitudes of these four main frequencies are multiples of each other.

[0038] Referring to step (d), in order to capture the time-varying characteristics of the food processor's vibration signal, it is necessary to perform time-frequency domain analysis on the non-stationary signal. Again, time-frequency domain processing is performed on a small segment of the five-cycle vibration signal from step (c). Time-domain, frequency-domain, and time-frequency domain analyses all involve analyzing a small segment of the effective time period of the original vibration signal. The STFT (Short-Time Fourier Transform) time-frequency analysis method is introduced, and the time and frequency domains of the signal are linked through the STFT. This transform is defined as:

[0039]

[0040] See Figure 4Comparing the time-frequency diagrams of the signals under the three operating conditions, from the overall time-spectrum perspective, the energy amplitude is highest when grinding solids, followed by empty cups, and lowest when grinding liquids. The highest energy in an empty cup is greater than the highest energy in the other two groups. Moreover, the energy amplitude in the presence of ice exhibits non-periodic changes over time, even displaying various random harmonics. This indicates that the blade rotation speed is irregular in a local time frame when the food processor is crushing ice, and there may even be instances where the ice cubes are spinning upwards and hitting empty air, resulting in irregular changes in the energy amplitude of the vibration signal.

[0041] From a temporal perspective, the frequency components in both the empty and full cups remained almost unchanged after the machine started. This is because the blade resistance was nearly constant under these two conditions, causing the blender to maintain a uniform stirring speed and thus the vibration frequency remained constant. Furthermore, the time-frequency graph for the third condition shows the changes in the state of the food throughout the process of crushing the ice. In the first 25 seconds after the machine started, the frequency gradually increased, indicating that as the ice was broken into smaller pieces, the blade's rotational resistance decreased, and the rotation speed increased accordingly. A significant increase in frequency occurred after 25 seconds, indicating that the ice had been broken into ice crystals and was eventually ground into water. Therefore, the frequency changes over time on the time-frequency graph can be used to determine the changes in the state of the food during the grinding process.

[0042] Analyzing the frequency distribution, we can see energy ridges around 85Hz, 170Hz, 255Hz, and 340Hz in the empty cup state. The lowest frequency component clearly has the highest energy ridge intensity, and the other two conditions are similar. This is because the food processor blades are four-bladed, resulting in vibration signals with frequency components ranging from 1st to 4th harmonics. If the 4th harmonic frequency components remain unchanged after the machine starts, it indicates that the ingredients have reached a steady state and have been ground into a liquid paste, or that the cup is empty. Selecting four frequency bands near the 4th harmonic of the vibration signal, we calculated the maximum frequency value and average amplitude for all samples. Comparing box plots under each condition reveals that the amplitude of the maximum frequency within each of the four frequency bands is more sensitive to determining whether solids are being stirred, while the average amplitude within the four frequency bands is less sensitive. Specifically, the highest frequency around 350Hz indicates an empty cup state, and the highest frequency around 330Hz indicates a liquid state. The fourth harmonic frequency component changes continuously over time, representing the solid grinding process. The abrupt change in the highest frequency around 320Hz indicates that the solid particles have been agitated into a paste. This confirms the conclusion of the previous time-frequency domain analysis, which suggests that the grinding conditions of a food processor can be determined by the information of the highest frequency component, which has higher sensitivity in the fourth harmonic frequency range.

[0043] Therefore, by utilizing the time-frequency domain characteristics of vibration signals, it is possible to make a simple and preliminary judgment on the type and degree of grinding of ingredients in the food processor.

[0044] Referring to step (e), all data under the three operating conditions are uniformly segmented and sampled using a sliding window. For the existing 60,000 data points for each operating condition, each sample takes data for ten cycles, i.e., 13,440 data points. After each half-cycle (672 data points) of sliding windowing, a sample is taken again, with 50 sets of data taken for each operating condition, for a total of 150 samples. All of these are used as the training set. More samples need to be taken as the test set, using the same sliding window method, and taking data of the same length as the training set samples (training set sample size: test set sample size = 5:2, consistent across all operating conditions). Another 20 sets of data are taken for each operating condition as the test set for identification and verification. Using this sliding window method, 60,000 data points are sufficient to obtain exactly 70 sets of samples. Furthermore, the subsequent calculations of the 16 indicators are performed separately on all samples in the training set, i.e., the first 50 samples for each operating condition are calculated separately.

[0045] For the 150 samples obtained in step (e), the vibration data of each sample are then uniformly segmented around the first to fourth harmonic frequency components, specifically 60-139Hz, 140-219Hz, 220-299Hz, and 300-379Hz. The maximum frequency value within each frequency band is calculated for each sample; see [reference needed]. Figure 8 Box plots are used for comparative analysis of indicators that are highly sensitive to distinguishing different working conditions.

[0046] By comparing the box plots under various operating conditions, it can be found that the maximum frequency values ​​differ significantly in the first, second, and fourth frequency bands, making them highly sensitive to determining whether solids are being stirred. In the 60-139Hz range, the maximum frequency is concentrated at 95Hz for an empty cup, at 85Hz for a full cup, and at 90Hz for a cup with ice. In the 140-219Hz range, the maximum frequency is concentrated at 195Hz for an empty cup, at 175Hz for a full cup, and at 190Hz for a cup with ice. In the 300-379Hz range, the maximum frequency is concentrated at 325Hz for an empty cup, at 335Hz for a full cup, and at 350Hz for a cup with ice.

[0047] Calculate the average amplitude within each frequency band under the three operating conditions, see [reference]. Figure 9 Box plots are used for comparative analysis of indicators that are highly sensitive to distinguishing different working conditions.

[0048] By comparing the box plots under various operating conditions, it can be found that the average amplitude in the four frequency bands is not very sensitive to the determination of whether there is solid in the stirring, and the differentiation under each operating condition is not ideal.

[0049] Referring to step (f), based on all sample data under different operating conditions obtained through sliding window sampling in step (e), the next step involves calculating and comparing multiple time-domain and frequency-domain characteristic parameters of the food processor vibration data to select parameters with significantly different sensitivity as effective feature indicators. Considering the characteristics of the food processor vibration signal, the selected time-domain characteristic parameters are mainly as follows:

[0050] (1) Variance reflects the degree of change in the amplitude of a signal over different times. Its formula is:

[0051]

[0052] (2) Peak factor, representing the extreme degree of the peak value in the waveform, exhibits good sensitivity to internal abnormal instantaneous impact vibrations. Its formula is:

[0053]

[0054] (3) The impulse factor reflects the waveform characteristics of a signal, that is, whether the signal exhibits significant transient changes. Its formula is:

[0055]

[0056] (4) Kurtosis, which reflects the shape of the signal amplitude distribution, i.e., the comparison between the overall rise and fall rates of the signal amplitude. Its formula is:

[0057]

[0058] To detect whether there are solid components in the target object inside the food processor, quantitative indicators that can reflect the signal vibration energy and impact characteristics are selected. The vibration data are calculated and compared using the above four time-domain signal characteristic parameters. Box plots are used for comparative analysis to distinguish indicators with high sensitivity to different working conditions.

[0059] See Figure 10Comparing the four time-domain characteristic parameters above, it is clear that the peak factor and impulse factor are more sensitive to determining whether ice is present during stirring. In an empty cup, the peak factor is mainly concentrated around 5; in a full cup, it is mainly concentrated around 5.5; and in the case of ice, it is mainly concentrated around 5.8. Similarly, in an empty cup, the impulse factor is mainly concentrated around 5.5; in a full cup, it is mainly concentrated around 7; and in the case of ice, it is mainly concentrated around 7.5. However, variance and kurtosis have relatively low discriminative power and sensitivity in determining whether ice is present during stirring.

[0060] Based on the characteristics of the experimental data, the selected frequency domain feature parameters mainly fall into the following categories:

[0061] (1) The mean spectral amplitude can reflect the energy distribution of the signal in the frequency domain. Its formula is as follows:

[0062]

[0063] (2) Spectral energy can reflect the total energy of a signal in the frequency domain, and its formula is as follows:

[0064]

[0065] (3) The centroid frequency can reflect the dominant frequency or the location of energy concentration of the signal in the frequency domain. Its formula is as follows:

[0066]

[0067] (4) Frequency variance reflects the degree of dispersion of the energy distribution of a signal in the frequency domain. Its formula is as follows:

[0068]

[0069] To address the specific characteristics of target detection regarding the presence or absence of solids, from an energy perspective, the above four characteristic parameters are selected for comparative analysis of the frequency domain signals of vibration data. (See also...) Figure 11 Box plots are used for comparative analysis of indicators that are highly sensitive to distinguishing different working conditions.

[0070] Comparative analysis of four frequency domain characteristic parameters clearly shows that the center-of-gravity frequency and frequency variance are highly sensitive to determining the presence of ice during stirring. In an empty cup, the center-of-gravity frequency is mainly concentrated around 268 MHz; in a full cup, it is mainly concentrated around 252 MHz; and with ice, it is mainly concentrated around 262 MHz. The frequency variance is mainly concentrated around 9500 MHz in an empty cup, around 9200 MHz in a full cup, and around 8500 MHz with ice. However, the energy levels in an empty cup and with ice are relatively similar, indicating that both the mean spectral amplitude and spectral energy are less sensitive to determining the presence of ice during stirring, and cannot be fully distinguished from the perspective of vibrational energy.

[0071] Therefore, based on the comparative analysis of the above 16 characteristic indicators (maximum frequency in four frequency bands, average amplitude in four frequency bands, four typical time-domain characteristic parameters, and four typical frequency-domain characteristic parameters), it was found that relying on only one or two signal indicators is insufficient to accurately distinguish the state of grinding. Therefore, it is necessary to package all indicators as training features and use a model for classification and recognition.

[0072] Referring to step (g), sort the samples according to the different working conditions they represent, i.e., the hardness level of the ingredients, and assign numerical labels to all samples in both the training and test sets. That is, all samples with ice are labeled "1", all samples with a full glass of water are labeled "2", and all samples with an empty glass are labeled "3". Use an SVM (Support Vector Machine) model, performing five cross-validations to prevent overfitting. Train the model using a Gaussian SVM. During training, the SVM model associates the input samples with their corresponding labels. Specifically, it distinguishes different working conditions based on the differences in 16 feature values ​​calculated above for different working conditions in the training set. Five cross-validations allow the model to better classify the training samples. This minimizes classification errors in the training samples. Observe the classification effect.

[0073] A relatively good recognition and classification effect can be obtained. See the obtained verification confusion matrix. Figure 5 The PPV values ​​on the main diagonal represent the positive prediction values ​​for each working condition. The accuracy rates for the three working condition classifications are 99.4%, 100%, and 100%, respectively. After multiple training iterations, the overall accuracy of the identification and classification remains above 99%. Therefore, using the SVM machine learning model can provide a preliminary identification and classification of the state of the target object inside the food processor.

[0074] Referring to step (h), based on the excellent classification results obtained in step (f), using all samples from the test set in step (e), each sample data was processed by a Gaussian SVM model for identification. The results showed that the label numbers output from the test set data of 60 samples across three working conditions were all correct, meaning the identification accuracy was 100%. Therefore, the model can be used to identify a segment of vibration signal data to obtain a label number, thereby accurately determining which food's state corresponds to the current grinding state of the food processor.

[0075] Example 2:

[0076] To further verify the effectiveness of using vibration signal characteristics to classify ingredients of different hardness processed by a food processor, six additional processing conditions with varying sample sizes were added to the original three processing conditions. Specifically, these included 166 samples of fresh vegetables, 153 samples of frozen antioxidants, 156 samples of Oreo milkshakes, 305 samples of nut butter, 116 samples of protein powder, 159 samples of smoothies, and other ingredients in different states, as well as 264 samples of ice cubes, 194 samples of full cups of water, and 57 samples of empty cups, for a total of nine processing conditions. Steps (a) to (f) in Example 2 are the same as in Example 1, except that the three processing conditions are replaced with nine. An SVM model is then used for training, and the classification results are observed after training.

[0077] The final training accuracy still reaches 99.2%, and the resulting validation confusion matrix can be found here. Figure 6 This demonstrates that using vibration signal characteristic indicators has a good classification effect on identifying different types of ingredients inside a food processor.

[0078] Referring to step (g), based on the excellent classification results above, the nine working conditions are classified according to the degree of solid grinding. See Table 1. The nine working conditions are classified in the following order: ice cubes, fresh green vegetables, frozen antioxidants, smoothies, protein powder, Oreo milkshakes, nut butter, full cup of water and empty cup, into large particle solids, small particle solids, solid-liquid mixtures, liquids and gases. Specifically, ice cubes represent large solid particles with a diameter of 10-100 mm; fresh green vegetables represent large solid particles with a diameter of 3-10 mm; frozen antioxidants represent small solid particles with a diameter of 1-3 mm; smoothies represent small solid particles with a diameter of 0.1-1 mm; protein powder represents small solid particles with a diameter of 10-100 μm; Oreo milkshakes represent solid-liquid mixtures with a viscosity of 100-500 Pa·s; nut butter represents solid-liquid mixtures with a viscosity of 10-100 Pa·s; a full glass of water represents a liquid with a viscosity of 0.1-1 m Pa·s; and an empty glass represents a gas with a viscosity of 15-20 μ Pa·s. Labels 1 to 9 are used sequentially. Detailed label numbers for different operating conditions can be found in Table 1 below.

[0079] Table 1.

[0080]

[0081] Referring to step (h), select different operating conditions, and for each condition, select 20 samples of vibration data, which are the test set samples. Based on the 16 feature indicators mentioned above, train and identify them separately. Finally, the label accuracy for the identified food state of the 180 samples is 100%, fully verifying the reliability of the selected features and model classification. Therefore, based on this, it is possible to accurately determine which type of food state corresponds to the current grinding state of the food processor. (Thus, using the above method, for example, if the identification result of the food processor's operating vibration data is label "3", it can be determined that the food being ground inside the food processor is small solid particles with a particle diameter of 1-3mm. If the identification result is label "9", it can be determined that the food being ground inside the food processor is either in a state of upward rotation and emptying or is already in an empty cup state.)

[0082] In summary, by using calculated values ​​based on 16 feature parameters as features and training with SVM, accurate identification and classification can be performed to determine the state of the food inside the machine. The nine operating conditions are categorized according to the degree of solid grinding (in the above order): large solids, easily breakable solids, small particles, viscous liquids, pure liquids, and empty cups, labeled 1 to 9 respectively. This allows for accurate identification of the grinding state corresponding to the current food processor's vibration signal processing.

[0083] This method can monitor the grinding status of ingredients inside the food processor in real time, while preventing the machine from continuously grinding without food and allowing for reasonable control of the processor's start-stop time. These studies will provide important references and guidance for the further optimized design of blending appliances such as juicers, soy milk makers, and high-speed blenders.

[0084] It is understood that the present invention has been described through some embodiments, and those skilled in the art will recognize that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the invention. Furthermore, under the teachings of the present invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the invention. Therefore, the present invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of the present invention.

Claims

1. A method for identifying the grinding state of a food processor based on vibration signal analysis, characterized in that, Includes the following steps, And the following steps are performed in sequence: Step (a): Use a micro accelerometer to collect the original vibration signals of the food processor under different operating conditions. Different operating conditions refer to the different hardness of the food being ground inside the food processor. Select a location where the micro accelerometer can obtain high data quality and use the micro accelerometer to collect vibration signal data of the food processor under different operating conditions. Step (b): Select vibration data of the x-axis from the data of the three directions of the micro accelerometer, and perform sliding window processing on the raw vibration signal data collected in step (a) to select the effective time period of the vibration signal; Step (c): For the non-stationary signal generated by the vibration of the food processor during operation, the characteristics of the time domain and frequency domain of a small segment of vibration signal in the effective time period under different working conditions are compared and analyzed by fast Fourier transform, as well as the different frequency components and the different distribution of signal energy in the frequency domain. Step (d): Based on step (c), perform a short-time Fourier transform on a small segment of the vibration signal from step (c) under different working conditions to obtain information in the time-frequency domain. Step (e) divides the vibration signal data of the effective time period under different working conditions obtained in step (b) into segments, and performs sliding window sampling with data of fixed length. The ratio of training set samples to test set samples is 5:2 for each working condition. Then, the data is evenly divided into segments centered on the four harmonics of the frequency domain information obtained in step (c), and the maximum frequency value and average amplitude in the four frequency bands of each segment of sample data in the training set are calculated respectively. Step (f) calculates eight feature functions for all sample data under different working conditions obtained by sliding window sampling in step (e) in the training set. The eight feature functions include four typical time-domain feature parameters and four typical frequency-domain feature parameters. The four typical time-domain feature parameters are: variance, peak factor, impulse factor, and kurtosis. The four typical frequency-domain feature parameters are: mean spectral amplitude, spectral energy, centroid frequency, and frequency variance. There are a total of eight typical time-domain and frequency-domain feature functions. Step (g): Use all 16 feature indicators—the maximum frequency in the four frequency bands, the average amplitude in the four frequency bands, four typical time-domain feature parameters, and four typical frequency-domain feature parameters—as training features. Sort the samples according to the different working conditions they represent, i.e., the hardness level of the food, and assign them numerical labels. Label all samples in the training and test sets accordingly. Use the support vector machine model to train all the samples in the training set from step (e). Through training, associate the model with the labels and obtain a classification result for different working conditions. Step (h), Step (h), Based on the classification results trained in Step (g), use all samples of the test set in Step (e) to perform recognition on each sample data through the support vector machine model, and observe whether the recognition of each sample is correct; if a 100% recognition accuracy is obtained, the model can be used to identify a segment of vibration signal data to obtain a label number, and then accurately determine which food's state corresponds to the current grinding state of the food processor.

Images