A raw material abnormality detection device and method based on plastic particle impact soundprint

Abstract

The application relates to the technical field of plastic raw material detection, and discloses a raw material anomaly detection device and method based on plastic particle impact voiceprints. The device comprises a feeding assembly, a falling material constraint assembly, an impact detection assembly, a signal acquisition assembly, an interference monitoring assembly and a processing assembly; by limiting the feeding parameters, falling path and impact conditions of the plastic particles, impact sound signals, impact vibration signals and background interference signals are collected, the effective impact period is determined based on the impact vibration signals, and interference suppression is performed on the effective impact sound signals; time domain features, frequency domain features and attenuation features are extracted, the voiceprint deviation degree is obtained by comparing the features with normal raw material voiceprint benchmarks, and then the raw material anomaly is determined. The application can realize rapid and continuous detection of plastic particle mixing, wrong material, moisture, impurity mixing, abnormal recycling material proportion and particle size specification.

Description

Technical Field

[0001] This invention relates to the field of detection technology, specifically to a raw material anomaly detection device and method based on the impact sound patterns of plastic particles. Background Technology

[0002] Plastic granules are a fundamental raw material in the production of plastic products. They are typically produced from polymers such as polyethylene, polypropylene, polyvinyl chloride, polyamide, and polycarbonate through a granulation process. During plastic molding processes such as injection molding, extrusion, blow molding, and calendering, factors such as the type of plastic granules, granule size, moisture content, filler ratio, modifier content, and mixing uniformity directly affect the mechanical properties, appearance quality, dimensional accuracy, and production stability of the plastic products. Therefore, it is crucial to inspect the raw materials for conditions such as mixing, incorrect materials, moisture, impurities, aged granules, abnormal proportions of recycled materials, or abnormal granule specifications before the plastic granules enter the production equipment.

[0003] Current methods for detecting anomalies in plastic raw materials mainly include manual visual inspection, density testing, melt flow index testing, moisture content testing, spectral analysis, and machine vision inspection. Among these, manual visual inspection relies heavily on operators' judgment of the color, transparency, shape, and impurities of plastic particles. The results are easily influenced by operator experience, ambient lighting, and fatigue levels, making it highly subjective and difficult to distinguish between plastic particles that look similar but have different material compositions. Density testing, melt flow index testing, and moisture content testing typically require sampling and specialized instruments, resulting in long testing cycles. Some tests also require heating, melting, or drying the samples, making it difficult to meet the needs of rapid, continuous testing on production sites.

[0004] Spectroscopic detection can identify raw material categories or compositional differences based on the optical response characteristics of different plastic materials. However, spectroscopic detection equipment is usually expensive and has certain requirements regarding the testing environment, particle surface cleanliness, and sampling location. When the plastic particles are covered with dust, have a dark color, or when the spectral characteristics of different materials are very similar, the accuracy of the detection may be affected. Machine vision inspection is mainly suitable for identifying visible anomalies such as color, particle size, shape, and surface defects. However, its detection capability is limited for anomalies that cannot be directly reflected by appearance, such as changes in material density, elastic modulus, internal structure, moisture content, and component ratios.

[0005] During the conveying, feeding, and discharging of plastic granules, impact sounds are generated when the granules collide with hoppers, guide plates, detection plates, or other structures. Because different types, formulations, or states of plastic granules vary in density, hardness, elasticity, damping characteristics, particle size, and surface morphology, the acoustic signals generated during their impacts will also differ in terms of time-domain waveform, frequency distribution, energy attenuation, impact duration, and acoustic signature characteristics. Therefore, utilizing the acoustic signature information generated by plastic granule impacts to identify the state of raw materials provides a non-contact, rapid, and easily implemented online technical approach for detecting anomalies in plastic raw materials.

[0006] However, existing technologies for material detection using acoustic signals are mostly applied to scenarios such as mechanical equipment fault diagnosis, bulk material flow monitoring, or particle size analysis, with few solutions specifically targeting anomaly detection in plastic granule raw materials. In actual production environments, the impact sound of plastic granules is easily affected by factors such as particle falling height, impact angle, particle quantity, conveying speed, impact structure material, equipment vibration, and environmental noise, resulting in highly random and non-stationary acoustic signals. When multiple plastic granules continuously or simultaneously impact the detection structure, the different impact sounds may also superimpose, making it difficult to accurately extract effective acoustic signature features.

[0007] Furthermore, different batches of normal plastic granules may exhibit variations in acoustic signature due to fluctuations in production processes, granule size, and additive content, while the acoustic signature differences between some abnormal raw materials and normal raw materials may be relatively small. If only sound pressure level, single frequency peak value, or fixed thresholds are used for judgment, misjudgments or omissions are likely, making it difficult to ensure accuracy across different raw material types and production conditions. Simultaneously, existing detection methods typically lack unified control over the plastic granule's falling state, impact conditions, and acoustic signal acquisition location, resulting in poor consistency and comparability of acoustic signature samples, which is detrimental to establishing a stable and reliable raw material anomaly determination model.

[0008] Therefore, how to provide a raw material anomaly detection device and method based on the impact acoustic signature of plastic particles, acquire the impact acoustic signal of plastic particles under relatively stable impact conditions, reduce the influence of environmental noise, equipment vibration, particle flow rate and random collision on the detection results, and extract effective acoustic signature features from the impact acoustic signal that can characterize the material type, physical state or compositional changes of plastic particles, so as to achieve rapid and accurate detection of mixed materials, wrong materials, moisture, impurities, abnormal proportion of recycled materials and abnormal particle size, etc., is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0009] Based on the above description, the present invention provides a raw material anomaly detection device and method based on the impact sound pattern of plastic particles. By constraining the feeding process, falling path and impact conditions of plastic particles, and combining the impact vibration signal to determine the effective impact period, the impact sound signal is suppressed by the background interference signal, thereby obtaining a target impact sound signal with a high signal-to-noise ratio. Furthermore, the time domain features, frequency domain features and attenuation features in the target impact sound signal are extracted and compared with the sound pattern benchmark of normal raw materials to realize the detection of anomalies in plastic particle raw materials.

[0010] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: A raw material anomaly detection device based on the impact sound of plastic particles, comprising:

[0011] Feeding assembly: used to receive the plastic particles to be tested and output the plastic particles to be tested according to preset feeding parameters;

[0012] Material dropping constraint component: disposed on the discharge side of the feeding component, used to limit the falling path of the plastic particles to be detected, so that the plastic particles to be detected are output according to preset impact conditions;

[0013] Impact detection assembly: includes an impactor and a support structure for supporting the impactor. The impactor is disposed below the material discharge restraint assembly and is used to receive the plastic particles to be tested output by the material discharge restraint assembly and generate an impact response when the plastic particles to be tested impact the impactor.

[0014] Signal acquisition component: includes an acoustic signal acquisition unit and a vibration signal acquisition unit. The acoustic signal acquisition unit is used to acquire the impact acoustic signal generated when the plastic particle to be detected impacts the impactor. The vibration signal acquisition unit is connected to the impact detection component and is used to acquire the impact vibration signal generated by the impactor.

[0015] Interference monitoring component: used to collect background interference signals in the environment where the impact detection component is located;

[0016] Processing component: Connected to the feeding component, signal acquisition component, and interference monitoring component respectively, the processing component is configured to: determine the effective impact period based on the impact vibration signal; extract the corresponding effective impact sound signal from the impact sound signal according to the effective impact period; perform interference suppression processing on the effective impact sound signal based on the background interference signal to obtain the target impact sound signal; extract impact acoustic features from the target impact sound signal, including at least time-domain features, frequency-domain features, and attenuation features; and compare the impact acoustic features with a pre-established normal raw material acoustic feature benchmark to obtain the acoustic feature deviation degree of the plastic particle to be detected.

[0017] When the degree of deviation of the voiceprint meets the preset abnormality judgment condition, the raw material abnormality detection result is generated.

[0018] Based on the above technical solution, the present invention can be further improved as follows.

[0019] Furthermore, the feeding assembly includes a storage bin, a quantitative discharging mechanism, and a discharging control mechanism; the quantitative discharging mechanism is located at the discharge port of the storage bin and is used to output a preset quantity or preset mass of plastic particles to be tested at preset time intervals; the discharging control mechanism is connected to the processing assembly and is used to adjust the discharging frequency and discharging amount of the quantitative discharging mechanism according to the feeding control command output by the processing assembly so that adjacent particle collision events are separated from each other in time.

[0020] Furthermore, the material dropping constraint component includes a vertically extending material dropping channel and an attitude rectification structure disposed within the material dropping channel. The inlet end of the material dropping channel is connected to the outlet side of the feeding component, and the outlet end of the material dropping channel faces the impact member. The attitude rectification structure is used to reduce the lateral motion component of the plastic particles to be detected and to assist the plastic particles to be detected in falling from a preset height range to the preset impact area of ​​the impact member.

[0021] Furthermore, the impactor is detachably mounted on the support structure, the impactor has an impact surface for impact by the plastic particles to be tested, a vibration isolation element is provided between the support structure and the mounting base, the vibration isolation element is used to suppress the transmission of mechanical vibration of the mounting base to the impactor, and at least one of the material, thickness and effective impact area of ​​the impactor is preset according to the type of raw material of the plastic particles to be tested.

[0022] Furthermore, the acoustic signal acquisition unit includes at least one main acoustic sensor, which is disposed to the side or below the impactor and spaced apart from the impactor. The vibration signal acquisition unit includes a vibration sensor fixed to the impactor or the support structure. The interference monitoring component includes a reference acoustic sensor, which is disposed outside the impact detection component and away from the preset impact area, such that the environmental noise component in the background interference signal acquired by the reference acoustic sensor is greater than the impact sound component.

[0023] Furthermore, the processing component is configured to: filter the impact vibration signal to obtain a vibration envelope signal;

[0024] The impact trigger time is determined when the vibration envelope signal exceeds a preset vibration trigger threshold.

[0025] Based on the impact trigger time, an effective impact period is constructed according to a preset forward time length and a preset backward time length;

[0026] When two adjacent valid impact periods overlap, the corresponding impact events are marked as superimposed impact events, and the valid impact sound signals corresponding to the superimposed impact events are removed or the valid impact sound signals corresponding to the superimposed impact events are processed by sound source separation.

[0027] Furthermore, the time-domain features include at least two of the following: peak sound pressure level, root mean square value, waveform factor, impulse factor, kurtosis, and duration; the frequency-domain features include at least two of the following: spectral centroid, main peak frequency, band energy ratio, spectral entropy, and Mel frequency cepstral coefficient; and the attenuation features include at least one of the following: envelope attenuation coefficient, reverberation attenuation time, adjacent peak attenuation ratio, and the time corresponding to the energy decreasing to a preset proportion.

[0028] Furthermore, the normal raw material acoustic signature benchmark includes multiple normal acoustic signature samples corresponding to normal plastic granules under different feeding parameters and environmental conditions. The processing component is configured to determine the benchmark value, allowable fluctuation range, and feature weight of each impact acoustic signature feature based on the multiple normal acoustic signature samples, and calculate the acoustic signature deviation degree based on the deviation of the impact acoustic signature feature relative to the benchmark value and the feature weight. The preset anomaly judgment condition includes at least one of the following conditions: the acoustic signature deviation degree is greater than a preset deviation threshold; the number of impact events with an acoustic signature deviation degree greater than the preset deviation threshold in a preset number of consecutive impact events reaches a preset number; the anomaly probability output by the pre-trained anomaly recognition model is greater than a preset probability threshold.

[0029] Furthermore, a method for detecting raw material anomalies based on the impact sound of plastic particles, employing the aforementioned raw material anomaly detection device based on the impact sound of plastic particles, includes:

[0030] The system outputs the plastic particles to be tested according to the preset feeding parameters and limits the falling path of the plastic particles to be tested so that the plastic particles to be tested impact the impactor according to the preset impact conditions.

[0031] The system collects the impact sound signal and impact vibration signal generated when the plastic particle to be tested impacts the impactor, and also collects the background interference signal in the environment where the impactor is located.

[0032] The effective impact period is determined based on the impact vibration signal;

[0033] The corresponding valid impact sound signal is extracted from the impact sound signal according to the effective impact time period;

[0034] Based on the background interference signal, the effective impact sound signal is subjected to interference suppression processing to obtain the target impact sound signal;

[0035] Extract impact acoustic signature features from the target impact acoustic signal, including at least time-domain features, frequency-domain features, and attenuation features;

[0036] The impact acoustic signature is compared with a pre-established acoustic signature benchmark for normal raw materials to obtain the degree of acoustic signature deviation of the plastic particles to be tested.

[0037] When the degree of deviation of the voiceprint meets the preset abnormality judgment condition, the raw material abnormality detection result is generated.

[0038] Furthermore, the method also includes: obtaining the raw material identifier corresponding to the plastic particles to be detected and calling the corresponding normal raw material acoustic signature based on the raw material identifier;

[0039] The corresponding voiceprint deviation degree is obtained based on a preset number of consecutive impact events;

[0040] The deviation of voiceprints corresponding to the preset number of consecutive impact events is statistically analyzed to obtain at least one of the following: the proportion of abnormal events, the average deviation, and the dispersion of deviation.

[0041] The abnormality type and abnormality level of the plastic particles to be tested are determined based on at least one of the abnormal event proportion, average deviation degree, and deviation degree dispersion; wherein, the abnormality type includes at least one of the following: mixing abnormality, incorrect material abnormality, moisture abnormality, impurity contamination abnormality, recycled material ratio abnormality, and particle size abnormality.

[0042] Compared with the prior art, the technical solution of this application has the following beneficial technical effects:

[0043] 1. The impact sound signal generated when plastic particles collide with impactors is used as information to characterize the state of raw materials. This eliminates the need for complex chemical treatment or destructive sampling of the plastic particles to be tested, which is conducive to the rapid and continuous detection of plastic particle raw materials.

[0044] 2. By controlling the discharge frequency and discharge amount of plastic granules through the feeding component, the collision events of adjacent granules are separated in time, which can reduce the superposition of acoustic signals caused by multiple granules colliding at the same time or in succession, and improve the extractability of acoustic features of a single impact.

[0045] 3. By limiting the falling path, falling height and impact area of ​​plastic particles through the material feeding channel and attitude rectification structure, the difference in sound signature caused by the lateral movement of particles, attitude change and impact position change can be reduced, and the comparability between different impact events can be improved.

[0046] 4. Simultaneously collecting impact sound signals and impact vibration signals, and using the impact vibration signals to determine the effective impact period, can make the triggering basis of the impact event correspond to the actual mechanical response of the impacting component, reducing the possibility of false triggering or missed triggering caused by environmental noise.

[0047] 5. An interference monitoring component is set up to collect background interference signals, and interference suppression processing is performed on the effective impact sound signal based on the background interference signals, which can reduce the impact of equipment noise, airflow noise and other background noise in the production environment on the target impact sound signal.

[0048] 6. By comprehensively extracting the time-domain, frequency-domain, and attenuation characteristics of the impact sound signal, it is possible to characterize the material properties and state changes of plastic particles from multiple dimensions such as impact intensity, spectral distribution, and energy attenuation process. Compared with detection methods that only use a single sound pressure amplitude or a single frequency parameter, it has richer evidence for anomaly identification.

[0049] 7. By establishing multiple normal acoustic fingerprint samples under different feeding parameters and environmental conditions, the baseline value, allowable fluctuation range and feature weight of each impact acoustic fingerprint feature can be determined, which can improve the adaptability of the normal raw material acoustic fingerprint baseline to changes in actual production conditions.

[0050] 8. By statistically analyzing the degree of deviation of voiceprints from multiple consecutive impact events, and combining the proportion of abnormal events, the average degree of deviation, and the dispersion of deviation, the type and level of abnormality can be determined. This can reduce misjudgments caused by single accidental impact differences and improve the stability of raw material anomaly detection results.

[0051] 9. It can be used to identify one or more of the following abnormalities: mixing, incorrect material, dampness, impurity contamination, abnormal recycled material ratio, and abnormal particle size, providing a basis for raw material quality control, incoming material inspection, and online abnormality early warning in the plastic product manufacturing process.

[0052] 10. The impactor adopts a detachable structure, and the material, thickness and effective impact area of ​​the impactor can be set according to the raw material type of the plastic particles to be tested. Therefore, it can adapt to different types, sizes and acoustic properties of plastic particles, and has good applicability and expandability. Attached Figure Description

[0053] Figure 1 This is a schematic diagram of the overall structure of the raw material anomaly detection device of the present invention;

[0054] Figure 2A This is a cross-sectional view of the feeding assembly in this invention;

[0055] Figure 2B This is a cross-sectional view of the material feeding channel;

[0056] Figure 2C For the material discharge channel along Figure 2A Sectional view at point AA;

[0057] Figure 3 Flowchart for signal acquisition and background interference monitoring setup;

[0058] Figure 4 This is a block diagram of the electrical connections and functional modules of the device;

[0059] Figure 5A A schematic diagram illustrating the determination of a single effective impact event;

[0060] Figure 5B A schematic diagram for determining a superimposed impact event where two effective impact periods overlap;

[0061] Figure 6 This is a schematic diagram of the voiceprint feature extraction and anomaly detection process.

[0062] Figure 7 This is a flowchart of the raw material anomaly detection method.

[0063] Reference numerals: 111, storage bin; 112, quantitative discharge mechanism; 113, discharge control mechanism; 114, discharge port; 121, material drop channel; 122, attitude rectification structure; 123, feed end; 124, discharge end; 131, impact component; 133, support structure; 134, vibration isolation component; 135, mounting base; 137, preset impact area; 142, main acoustic sensor; 144, vibration sensor; 151, reference acoustic sensor; 160, processing component; 171, particle collection container; 200, plastic particles to be detected. Detailed Implementation

[0064] To facilitate understanding of this application, a more complete description will be provided below with reference to the accompanying drawings, which illustrate embodiments of the present application. However, the present application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of this application will be thorough and complete.

[0065] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.

[0066] This invention targets polypropylene, polyethylene, polyamide, or polycarbonate plastic granules used in injection molding production lines as the objects to be tested. The plastic granules can be cylindrical, near-spherical, ellipsoidal, irregularly shaped, or granulated, with an equivalent particle size generally ranging from 1mm to 8mm and a single particle weight generally from 0.005g to 0.3g. This invention can also be used to test glass fiber reinforced plastic granules, mineral-filled plastic granules, flame-retardant plastic granules, recycled modified plastic granules, and plastic granules containing masterbatch.

[0067] The types of anomalies in this invention can include one or more of the following: mixing anomalies, incorrect material anomalies, moisture anomalies, impurity contamination anomalies, recycled material ratio anomalies, particle size anomalies, and aging particle anomalies. Specifically, mixing anomalies refer to the presence of plastic particles in the target plastic particles that differ from the target raw material type, formula, or batch; incorrect material anomalies refer to the actual type of raw material input being inconsistent with the type specified in the production task; moisture anomalies refer to the moisture content of the plastic particles exceeding the allowable range for the corresponding material; impurity contamination anomalies refer to the presence of metal shavings, sand, glass fragments, wood chips, lumps, dust clumps, or other foreign matter in the plastic particles; recycled material ratio anomalies refer to the actual proportion of recycled or regenerated materials in the raw materials deviating from the preset formula ratio; and particle size anomalies refer to a significant deviation from the normal range in particle size, length, cross-sectional dimensions, or particle shape.

[0068] A raw material anomaly detection device based on the impact sound of plastic particles, such as Figures 1-7As shown, the device includes a frame, feeding assembly, material feeding restraint assembly, impact detection assembly, signal acquisition assembly, interference monitoring assembly, processing assembly 160, power supply assembly, and optional abnormal material handling assembly. The frame supports the various functional components and can be made of cold-rolled steel plate with a thickness of 1.5mm-3mm by bending and welding, or assembled from industrial aluminum profiles with cross-sectional dimensions of 30mm×30mm, 40mm×40mm, or 45mm×45mm. The frame can be equipped with a closed or semi-closed casing, preferably made of stainless steel plate with a thickness of 1mm-2mm. The overall dimensions of the device are approximately 400mm-800mm wide. The device measures 160 mm in diameter, 350 mm to 700 mm in depth, and 800 mm to 1600 mm in height. The feeding assembly is located on the upper part of the frame, the material discharge restraint assembly is located below the feeding assembly, and the impact detection assembly is located below the discharge end 124 of the material discharge restraint assembly. Signal acquisition components are located near the impact detection assembly and on the impact detection assembly itself. Interference monitoring components are located outside the impact detection assembly and maintain a preset distance from the main impact area. The processing assembly 160 is located in an independent electrical control box. The device can be installed independently at the raw material testing station or connected upstream of a storage tank, dryer, vacuum feeder, weighing and batching machine, injection molding machine hopper, or extruder hopper. The device can use either bypass sampling detection or full-volume online detection. When using bypass sampling detection, a small amount of plastic particles is periodically taken from the main feed pipe or storage silo 111 and fed into the device. After testing, the sample is discharged into the qualified material channel, abnormal material channel, or recycling container. When using the full online detection method, plastic particles can pass through the device in batches or in continuous small flow rates, and the subsequent material distribution mechanism can be controlled according to the detection results. The feeding component is used to receive the plastic particles 200 to be detected and output the plastic particles 200 to be detected to the material discharge constraint component according to the preset feeding parameters. The feeding component includes a storage bin 111, a quantitative discharge mechanism 112 and a discharge control mechanism 113. The storage bin 111 is installed on the top of the frame and can be fixedly connected to the frame by bolts, clamps or welded brackets.

[0069] like Figures 1-7As shown, the storage silo 111 preferably has a funnel structure that is wider at the top and narrower at the bottom. Its upper cross-section can be circular, and its lower part forms a conical section that tapers towards the outlet 114. The effective volume of the storage silo 111 is 1L-50L, 2L-10L for laboratory or bypass testing devices, and 10L-30L for online testing devices. The storage silo 111 is made of 304 stainless steel, and the surface roughness of the inner wall of the storage silo 111 is less than or equal to Ra1.6μm to reduce plastic particle adhesion and residue. For devices requiring observation of the internal state, the entire silo is made of transparent polycarbonate. An inlet is located at the top of the storage silo 111, and an openable silo cover is provided at the inlet. A silicone rubber sealing ring is installed between the silo cover and the silo body to reduce external impact. Dust and noise enter the storage silo 111. A discharge port 114 is provided at the bottom of the storage silo 111. The inner diameter of the discharge port 114 is 8mm-50mm, which is determined according to the particle size and the feeding amount. To prevent particle bridging, a micro vibrator can be installed on the outer wall of the storage silo 111. The micro vibrator preferably uses an eccentric vibration motor with a rated voltage of 24V and a rated power of 5W-30W. The processing component 160 only starts the vibrator when it detects insufficient discharge or bridging in the storage silo 111, so as to avoid the mechanical vibration of the vibrator during operation being transmitted to the impact detection component. The quantitative discharge mechanism 112 is set at the discharge port 114 of the storage silo 111. Its function is to output a preset quantity or preset mass of plastic particles according to a preset time interval.

[0070] like Figures 1-7 As shown, a quantitative discharging mechanism 112 adopts a rotary quantitative disc. The rotary quantitative disc includes a disc body, a plurality of metering holes evenly arranged along the circumference, a drive shaft and a reduction motor. The disc body is located below the discharge port 114 of the storage bin 111. An axial gap of 0.1mm-0.8mm is maintained between the disc body and the bottom of the storage bin 111. The diameter of the metering hole is 1.2-2.5 times the maximum outer diameter of the particle, and the depth is 1-3 times the average length of the particle. The geared motor can be a stepper motor, specifically a two-phase stepper motor with a rated voltage of 24V and a step angle of 1.8°, and equipped with a planetary reducer with a reduction ratio of 5:1-20:1. The processing component 160 controls the rotation angle, rotation speed, and dwell time through a stepper driver. The discharge control mechanism 113 is electrically connected to the processing component 160 and includes a motor driver, a solenoid valve drive circuit, a relay, a photoelectric detector, a weighing sensor, and a control interface. The processing component 160 sends a feeding control command to the discharge control mechanism 113 according to the detection cycle. To ensure that two adjacent particle impact events are separated in time, the feeding interval should be greater than the main attenuation duration of a single impact sound signal. For the metal impactor 131, the duration of a single effective impact sound signal is usually 10ms-150ms, so the adjacent feeding interval can be set to 100ms-1000ms, typically 200ms-500ms.

[0071] like Figures 1-7 As shown, in single-particle detection mode, one plastic particle is output at a time, and the output time interval between adjacent particles is preferably 250ms-600ms. In small-batch detection mode, two to ten plastic particles are output at a time, and particles from the same batch collide with the impactor 131 in a short period of time. The comprehensive sound pattern formed by this batch is then used as the detection object. To ensure accurate feeding, a particle counting sensor can be installed below the quantitative discharging mechanism 112. The particle counting sensor can be a through-beam photoelectric sensor, a laser sensor, or a capacitive sensor. The through-beam beam passes through the discharging path, and when a particle passes by, it blocks the beam. The processing component 160 counts the number of particles based on this. In the method of quantitative feeding by mass, a weighing sensor is installed below the storage bin 111 or the quantitative container. The weighing sensor range can be 500g, 1kg, 2kg, or 5kg, and the comprehensive accuracy is preferably not less than 0.05%FS. The processing component 160 controls the discharging mechanism based on the weighing data to keep the mass of a single feeding between 0.05g and 10g.

[0072] like Figures 1-7 As shown, the material discharge constraint component is located on the discharge side of the feeding component, including a vertically extending material discharge channel 121 and an attitude rectification structure 122 located within the material discharge channel 121. The inlet end 123 of the material discharge channel 121 is connected to the discharge side of the feeding component, and the discharge end 124 faces the preset impact area 137 of the impactor 131. The material discharge channel 121 is used to limit the falling position, lateral velocity and impact angle of the plastic particles, so that different particles impact the impactor 131 under the most consistent conditions. The material discharge channel 121 is a closed channel formed by a circular tube. The feeding channel 121 is made of transparent polycarbonate. In one structure, the feeding channel 121 is a transparent polycarbonate round tube with an inner diameter of 8mm-25mm, a wall thickness of 1mm-3mm, and a total length of 150mm-800mm. For plastic granules with a particle size of 2mm-5mm, the inner diameter is 10mm-16mm. The inner diameter of the feeding channel 121 is generally set to 2-5 times the maximum outer diameter of the granules to be tested to avoid material jamming and to limit the lateral swing of the granules. The discharge end 124 of the feeding channel 121... The vertical distance between the particle and the impact surface of the impactor 131 constitutes the effective free fall height of the particle. The effective free fall height can be 100mm-1200mm, usually 250mm-700mm. For ordinary polypropylene, polyethylene or ABS particles, it can be set to 300mm-500mm. For particles with lighter weight or lower hardness, it can be appropriately increased to 500mm-800mm. For glass fiber reinforced, mineral filled or larger particles, it can be set to 200mm-400mm.

[0073] like Figures 1-7As shown, the dropping height can be adjusted by a telescopic tube, sliding guide rail, screw lifting mechanism, or multi-hole mounting plate. For example, the dropping channel 121 is fixed to the vertical guide rail by a pipe clamp. After loosening the locking bolt, it can move up and down. After adjustment, it can be re-locked. A height scale is set next to the guide rail, and the adjustment accuracy can reach 1mm-5mm. The attitude rectification structure 122 is used to reduce the lateral motion component of the plastic particles and guide the plastic particles to fall from the preset height range to the preset impact area 137 of the impactor 131. The attitude rectification structure 122 includes at least one constriction section, at least one set of guide plates, honeycomb guide holes, flexible bristles, flow divider baffles, or tapered guide sleeves. In one structure, the dropping channel 121 is arranged from top to bottom as a first-stage funnel-shaped constriction section, a vertical guide section, and a second-stage rectification section. The inlet diameter of the constriction section is 20mm-40mm, the outlet diameter is 8mm-15mm, the cone angle is 20°-60°, the length of the vertical guide section is 50mm-200mm, and its inner diameter is basically constant. The secondary rectification section is set 20mm-100mm above the discharge end 124, including 3-8 thin sheet-like guide ribs evenly arranged circumferentially. The thickness of the guide ribs can be 0.3mm-1.5mm, extending vertically, and their inner edges form a central area for particles to pass through. The guide ribs cause particles that have undergone lateral movement after colliding with the pipe wall to return to the center and fall. In another structure, the attitude rectification structure 122 adopts a honeycomb rectification component. The honeycomb rectification component has multiple vertical through holes, the diameter of each through hole is 1.2 times to 2 times the maximum outer diameter of the particle, and the through hole length is 20mm-80mm. The honeycomb rectifier is detachably inserted into the feeding channel 121 and can be replaced according to different particle sizes. A replaceable limiting nozzle can be installed at the discharge end 124 of the feeding channel 121. The inner diameter of the limiting nozzle's outlet is 4mm-15mm. The vertical deviation between the outlet center and the center of the preset impact area 137 of the impactor 131 is preferably no more than 2mm. The preset impact area 137 can be a circular area with a diameter of 10mm-80mm centered on the geometric center of the impactor 131, or a rectangular area with a length of 10mm-100mm and a width of 10mm-60mm. Preferably, the particle impact position is located at least 20mm inside the edge of the impactor 131 to reduce edge reflection and the influence of the mounting structure on the acoustic signature. A flexible connecting sleeve can be installed between the feeding channel 121 and the frame. The flexible connecting sleeve is made of soft PVC and its function is to reduce the transmission of vibration from the feeding motor or frame to the feeding channel 121 and the impact detection component.

[0074] like Figures 1-7As shown, the impact detection assembly includes an impactor 131 and a support structure 133 for supporting the impactor 131. The impactor 131 is located below the material discharge constraint assembly and is used to receive plastic particles output from the material discharge channel 121. It generates impact sound response and impact vibration response when the particles impact. The impactor 131 adopts a plate-like structure. To improve the consistency of the sound pattern, a plate-like impactor 131 with uniform thickness is used. In one measurement, the impactor 131 is a circular stainless steel plate with a diameter of 100mm-300mm and a thickness of 0.8mm-5mm. For ordinary plastic particles with a particle size of 2mm-5mm, a 304 stainless steel circular plate with a diameter of 180mm and a thickness of 1.5mm is used.

[0075] like Figures 1-7As shown, the impactor 131 can also be made of aluminum alloy plate, carbon steel plate, brass plate, ceramic plate, glass plate, polycarbonate plate, or composite material plate. Different materials have different natural frequencies, sound radiation efficiencies, and damping characteristics. Therefore, the impactor 131 can be selected according to the 200 types of plastic particles to be tested. For particles with low hardness and low impact energy, such as polyethylene and polypropylene, stainless steel plates or aluminum alloy plates with a thickness of 0.8mm-2mm are used to obtain higher sound signal amplitude. For glass fiber reinforced polyamide, polycarbonate, or mineral-filled particles, stainless steel plates or carbon steel plates with a thickness of 1.5mm-4mm can be used. For testing tasks that require highlighting particle damping differences, polycarbonate plates or epoxy composite plates with a thickness of 2mm-8mm can be used. The effective impact area of ​​the impactor 131 is determined according to the feed rate and impact point stability. When testing a single particle, the diameter of the effective impact area is preferably 20mm-60mm. When testing in small batches, the diameter of the effective impact area can be set to 50mm-120mm. The impactor 131 is detachably mounted on the support structure 133. Specifically, threaded connections, pressure ring connections, snap-fit ​​connections, magnetic connections, or clamping connections can be used. In one embodiment, the support structure 133 includes an annular support seat, an upper pressure ring, and multiple locking screws. The edge of the impact member 131 overlaps on the annular support seat, and the upper pressure ring presses against the edge of the impact member 131. The locking screws pass through the upper pressure ring in sequence and connect to the threaded holes on the annular support seat. A silicone rubber washer with a thickness of 0.5mm-2mm can be placed between the edge of the impact member 131 and the support seat to avoid rigid loosening and abnormal impact noise. The clamping width of the edge of the impact member 131 can be 5mm-20mm. To maintain consistent boundary conditions after different replacement operations, each locking screw can be tightened with a torque screwdriver to the same torque, for example, 0.5N·m-3N·m. The support structure 133 can be a three-point support, a four-point support, an annular continuous support, or a cantilever. For the support, a ring support or a three-point support evenly distributed along the circumference is preferred to ensure the stability of the vibration mode of the impactor 131. Vibration isolation components 134 are set between the support structure 133 and the mounting base 135. The mounting base 135 can be a frame beam, a mounting base plate, or an independent detection box. The vibration isolation components 134 are used to suppress the transmission of mechanical vibrations from the frame, the feeding motor, the conveying equipment, and the workshop floor to the impactor 131. The vibration isolation components 134 are made of solid rubber blocks with a Shore A hardness of 30-70. Each vibration isolation component 134 is set as a cylinder with a diameter of 15mm-40mm and a height of 10mm-30mm. 3-4 vibration isolation components 134 are evenly arranged below the support structure 133.

[0076] like Figures 1-7As shown, the upper end of the vibration isolator 134 is connected to the support structure 133 by bolts, and the lower end is connected to the mounting base 135 by bolts. In order to avoid vibration short circuits caused by the connecting bolts, rubber sleeves can be set around the bolt holes, or the bolts can only play a limiting role without completely compacting the vibration isolator 134. The natural frequency of the vibration isolation system is preferably lower than the main environmental mechanical vibration frequency. In practical applications, the natural frequency of the vibration isolation system composed of the support structure 133 and the impactor 131 can be selected to be 5Hz-30Hz, while the main frequency range of the impact sound pattern to be analyzed is 500Hz-20kHz, thereby reducing low-frequency mechanical vibration interference. A particle collection structure is set below or to the side of the impactor 131. The particle collection structure can include an inclined guide trough, an annular collection plate and a discharge port. After impact, the particles enter the qualified material container or sorting mechanism along the inclined guide chute to avoid the particles staying on the impactor 131 and causing secondary collisions. The impactor 131 is inclined at 2°-20° relative to the horizontal plane, usually 5°-10°, so that the particles slide away from the impact area by gravity after impact. When the impactor 131 is inclined, the axis of the discharge channel 121 should point to the preset impact area 137, and the position of the discharge end 124 can be corrected according to the inclination angle.

[0077] like Figures 1-7 As shown, the signal acquisition component includes an acoustic signal acquisition unit and a vibration signal acquisition unit. The acoustic signal acquisition unit is used to acquire the impact acoustic signal generated when plastic particles impact the impactor 131, and the vibration signal acquisition unit is used to acquire the impact vibration signal generated by the impactor 131. The acoustic signal acquisition unit includes at least one main acoustic sensor 142, using a pre-polarized measurement microphone with a sensitivity of 20mV / Pa-50mV / Pa and a frequency response range of 20Hz-20kHz. The main acoustic sensor 142 is disposed to the side or below the impactor 131, and spaced apart from the impactor 131 to avoid direct mechanical contact between the sensor and the impactor 131. When the main acoustic sensor 142 is disposed below the impactor 131, the main acoustic sensor 142... The sound hole faces the lower surface of the impactor 131. The vertical distance between the sound hole and the lower surface of the impactor 131 is 30mm-150mm, usually 50mm-100mm. The horizontal deviation between the central axis of the main sound sensor 142 and the center of the impact area is no more than 50mm. When the main sound sensor 142 is set to the side of the impactor 131, the straight distance between the sensor sound hole and the center of the impact area is 50mm-300mm. The angle between the sensor axis and the surface of the impactor 131 is 20°-80°. The main sound sensor 142 is fixed on an independent mounting base by a sensor bracket. The sensor bracket is made of stainless steel. A soft silicone rubber sleeve is set between the main sound sensor 142 and the bracket to reduce the transmission of bracket vibration to the microphone.

[0078] like Figures 1-7As shown, to improve noise immunity, two or more main acoustic sensors 142 can be set. For example, two main acoustic sensors 142 can be symmetrically arranged below the impactor 131 relative to the impact area. The processing component 160 can perform amplitude averaging, phase alignment, beamforming, or consistency verification on the two channels. An acoustic shield can be set on the outside of the main acoustic sensors 142. The acoustic shield uses a metal shell with a thickness of 1mm-3mm, and the inner wall is pasted with open-pore sound-absorbing sponge with a thickness of 5mm-30mm. The shield has a pickup hole on the side facing the impactor 131, and is as closed as possible in other directions, thus... To reduce ambient noise in the workshop, the acoustic signal acquisition channel includes a preamplifier, an anti-aliasing filter, and an analog-to-digital converter. When using an analog microphone for output, the preamplifier gain can be 10dB-40dB. The cutoff frequency of the anti-aliasing low-pass filter can be 18kHz-45kHz. The preferred resolution of the analog-to-digital converter is 16-bit, 24-bit, or higher. The acoustic signal sampling frequency can be 44.1kHz, 48kHz, 96kHz, or 192kHz. For devices that mainly analyze acoustic signatures below 20kHz, a sampling frequency of 48kHz or 96kHz and a quantization precision of 24-bit are typically used.

[0079] like Figures 1-7 As shown, the vibration signal acquisition unit includes a vibration sensor 144 fixed to the impactor 131 or the support structure 133. The vibration sensor 144 is used to determine the actual impact moment of the particles and to assist in identifying superimposed impacts and non-particle interference. The vibration sensor 144 can be a MEMS accelerometer. The selected vibration sensor 144 is a triaxial MEMS accelerometer with a range of ±16g, ±32g, or ±100g, a bandwidth of not less than 1kHz, and a sampling frequency of 5kHz-50kHz. The vibration sensor 144 can be fixed to the lower surface of the impactor 131 near the edge by threads, structural adhesive, or clamps. When it is not desired to increase the added mass of the impactor 131, the vibration sensor 144 can be fixed to the annular support. The vibration sensor 144 is mounted on the impactor 131, and the vibration trigger threshold is calibrated through experiments. When the vibration sensor 144 is fixed to the impactor 131, its mass does not exceed 5% of the mass of the impactor 131 to reduce the influence of the sensor on the natural frequency and vibration mode of the impactor 131. When structural adhesive is used for bonding, epoxy structural adhesive or cyanoacrylate adhesive can be used, and the adhesive layer thickness is controlled between 0.05mm and 0.3mm. When threaded connection is used, the bottom surface of the sensor and the impactor 131 should fit tightly together, and the connecting screws should be subjected to a stable preload. The vibration signal sampling frequency is an integer division of the acoustic signal sampling frequency or the same sampling clock is used. For example, when the acoustic signal sampling frequency is 48kHz, the vibration signal can be 12kHz, 24kHz, or 48kHz for time synchronization.

[0080] like Figures 1-7As shown, the main acoustic sensor 142, vibration sensor 144, and reference acoustic sensor 151 are preferably connected to the same multi-channel data acquisition board and synchronously acquired using the same sampling clock. The multi-channel data acquisition board can be a 4-channel or 8-channel, 24-bit, 48kHz-192kHz synchronous sampling audio acquisition board, or an industrial data acquisition card. The time deviation between different acquisition channels is no greater than one sampling period. When the acoustic signal and vibration signal use different acquisition modules, the processing component 160 can output a synchronous trigger signal or inject synchronous pulses into each channel for time alignment, with a time synchronization error of less than 0.1ms. The interference monitoring component is used to acquire background interference signals in the environment where the impact detection component is located. The interference monitoring component includes at least one reference acoustic sensor 151. The reference acoustic sensor 151 can be the same model as the main acoustic sensor 142, or it can be a microphone with a similar frequency response. The reference acoustic sensor 151 is located outside the impact detection assembly and away from the preset impact area 137, so that the ambient noise component in its collected signal is greater than the impact sound component. The reference acoustic sensor 151 is located outside the device housing and is 300mm-1500mm away from the center of the impact area, while the main acoustic sensor 142 is 50mm-200mm away from the center of the impact area. The sound hole of the reference acoustic sensor 151 faces the main ambient noise source. A sound insulation plate can be installed between the main acoustic sensor 142 and the reference acoustic sensor 151 to attenuate the particle impact sound when it reaches the reference acoustic sensor 151. In order for the reference acoustic sensor 151 to effectively collect the background noise of the fan, conveyor motor, vacuum feeder and workshop, the reference acoustic sensor 151 should not be completely placed in a sealed sound-absorbing structure. A steel plate with a thickness of 1mm-3mm can be installed between the reference acoustic sensor 151 and the impact component 131, and sound-absorbing cotton is installed on the inner side of the steel plate.

[0081] like Figures 1-7As shown, the processing component 160 can calculate the short-time cross-correlation between the main sound channel and the reference sound channel. When highly correlated continuous noise occurs simultaneously in both channels, it is identified as environmental interference. When the main sound channel experiences a significant transient impact while the reference sound channel has a weak response, it is identified as particle impact sound. In cases of strong environmental mechanical vibration, the interference monitoring component may also include a reference vibration sensor 144, which is fixed on the frame or mounting base 135 and used to collect vibrations of the equipment body. The processing component 160 can eliminate mechanical vibration interference caused by the feeding motor, fan, and conveying device based on the reference vibration signal. The processing component 160 is electrically connected to the feeding component, signal acquisition component, and interference monitoring component, respectively. The processing component 160 includes an industrial computer, embedded processor, digital signal processor, microcontroller, field-programmable gate array, or a combination thereof. In one embodiment, the processing component 160 includes an ARM processor, more than 4GB of RAM, more than 32GB of storage, a 24-bit multi-channel acquisition interface, a digital input / output interface, and an Ethernet communication interface. The processor's main frequency is preferably not less than 1GHz. The processing component 160 is connected to the quantitative discharging mechanism 112 through the motor drive interface, and to the electromagnetic gate, vibrator, and abnormal material distribution mechanism through the digital output interface. It is connected to the sound signal acquisition unit, vibration signal acquisition unit, and interference monitoring component through the analog or digital communication interface. The device is preferably powered by a 24V DC power supply. The sensors and processing component 160 can be powered separately by isolated DC-DC power supplies. The analog acquisition circuit and the motor drive circuit are wired in separate sections. The motor power line and the sensor signal line are kept at least 50mm apart, or they can be set in different cable trays. The sensor signal line is a shielded twisted pair cable or a shielded coaxial cable. The shielding layer is preferably grounded at a single point at the acquisition end. The frame and metal casing are connected to a protective ground. The analog ground, digital ground, and casing ground are connected according to electromagnetic compatibility requirements to reduce motor switching noise. The processing component 160 can be equipped with a touch screen for parameter setting and result display. The displayed content can include raw material identification, current feeding parameters, number of impacts, number of valid events, number of abnormal events, sound pattern deviation degree, abnormality type, abnormality level, and alarm status.

[0082] like Figures 1-7As shown, the processing component 160 first preprocesses the vibration signal. Vibration signal preprocessing may include DC removal, bandpass filtering, rectification, squaring, moving average root mean square calculation, and envelope extraction. A bandpass filter with a cutoff frequency of 100Hz-8000Hz is used for the vibration signal. For a vibration signal with a sampling frequency of 24kHz, a fourth-order Butterworth bandpass filter can be used, with a lower cutoff frequency of 200Hz and an upper cutoff frequency of 8000Hz. The absolute value of the filtered vibration signal is taken, and then the envelope is calculated using a sliding window with a length of 0.1ms-5ms. Hilbert transform can also be used to obtain the vibration envelope signal. The preset vibration trigger threshold can be a fixed threshold or a dynamic threshold. When using a fixed threshold, background vibration signals are first collected for 10-60 seconds under no-particle-impact conditions. The average value μv and standard deviation σv of the background envelope are calculated. The vibration trigger threshold Tv is then set as: Tv = μv + Kv × σv; where Kv is the threshold coefficient, which can be 3-10, preferably 5-8. When the number of sampling points where the vibration envelope signal continuously exceeds the preset vibration trigger threshold reaches a preset number, an impact is determined to have occurred. For example, when the sampling frequency is 24kHz, the envelope can be required to continuously exceed the threshold for 0.1ms-1ms to eliminate single-point electrical noise.

[0083] like Figures 1-7As shown, the impact trigger time can be defined as the moment when the vibration envelope first exceeds the threshold, or it can be defined as the moment corresponding to the local vibration peak. The processing component 160 uses the impact trigger time as a reference and constructs an effective impact period according to preset forward and backward time lengths. The forward time length is used to retain the background baseline before the impact, and the backward time length is used to fully cover the generation and attenuation process of the impact sound. The forward time length can be 1ms-30ms, preferably 5ms-15ms; the backward time length can be 20ms-500ms, preferably 50ms-200ms. For example, if the forward time length is set to 10ms and the backward time length is set to 120ms, then for the impact trigger time t0, the effective impact period is [t0-10ms, t0+120ms]. The processing component 160 extracts from the continuous impact sound signal collected by the main sound sensor 142 according to the effective impact period. For the corresponding valid impact sound signal, when the interval between two adjacent impact trigger times is less than the length of the valid impact time period, resulting in the overlap of the two valid impact time periods, the processing component 160 marks the corresponding event as a superimposed impact event. In embodiments with high requirements for detection stability, the superimposed impact event is directly discarded and not used for the calculation of the sound signature deviation degree. An instruction to reduce the feeding frequency is sent to the discharge control mechanism 113. For example, when more than 3 superimposed impacts occur in 20 consecutive events, the processing component 160 increases the feeding interval from 300ms to 400ms. If the superposition ratio still exceeds 10%, the feeding interval is increased until the superposition ratio drops to the set range. In embodiments that require improved detection efficiency, the superimposed impact sound signal can be separated into sound sources. The sound source separation can employ non-negative matrix factorization, independent component analysis, sparse decomposition, matching pursuit, or a transient sound separation algorithm based on neural networks.

[0084] like Figures 1-7As shown, the processing component 160 performs interference suppression on the effective impact sound signal based on the background interference signal collected by the reference acoustic sensor 151 to obtain the target impact sound signal. The interference suppression can employ spectral subtraction, adaptive filtering, Wiener filtering, reference noise cancellation, multi-channel beamforming, or a combination thereof. In one embodiment, a normalized least mean square adaptive filtering algorithm is used. The background interference signal collected by the reference acoustic sensor 151 is used as the input of the adaptive filter, and the signal collected by the main acoustic sensor 142 is used as the desired signal. The adaptive filter estimates the noise component formed by the propagation of environmental noise to the main acoustic sensor 142, and then subtracts the estimated noise from the main acoustic signal. The order of the adaptive filter can be 16-512, preferably 64-256; the step size factor can be 0.001-0.5. In another embodiment, short-time Fourier transforms are performed on the effective impact sound signal and the background interference signal, respectively. The window function uses a Hanning window with a frame length of 256-4096 points and a frame shift of 1 / 4-1 / 2 of the frame length. The noise power spectrum is estimated based on the background interference signal, and the power spectrum of the effective impact sound signal is subtracted from the noise power spectrum multiplied by a suppression coefficient. The suppression coefficient can be 0.5-2.5. To avoid generating musical noise, a minimum spectral floor is set for the subtraction result, with a spectral floor coefficient of 0.01-0.2. After interference suppression, the target impact sound signal is bandpass filtered. The bandpass range can be set to 200Hz-20kHz, preferably 500Hz-18kHz, depending on the impactor 131 and particle type. The processing component 160 can also normalize the amplitude of the target impact sound signal. However, when the peak sound pressure or absolute energy itself is used as a detection feature, the amplitude information should not be completely eliminated; the original amplitude characteristics can be retained, while a normalized signal is generated to extract the spectral shape and attenuation characteristics.

[0085] like Figures 1-7As shown, the processing component 160 extracts time-domain features, frequency-domain features, and attenuation features from the target impact sound signal. Time-domain features may include peak sound pressure level (PSL), root mean square (RMS) value, waveform factor, impulse factor, kurtosis, skewness, peak-to-peak value, energy, and duration. The PSL can be the maximum absolute value of the target impact sound signal. After microphone calibration, it can also be converted to Pascal or decibel sound pressure level. The RMS value is obtained by taking the square root of the average of the squares of each sampling point within the effective impact period. The waveform factor can be defined as the ratio of the RMS value to the average absolute value. The impulse factor can be defined as the ratio of the peak value to the average absolute value. Kurtosis reflects the impact and sharpness of the waveform. Duration can be defined as the duration from when the signal envelope first exceeds the energy threshold to when it falls below the threshold and remains below it for a preset time. The energy threshold can be set to 1%-20% of the peak envelope. Frequency-domain features may include spectral centroid, main peak frequency, secondary peak frequency, band energy ratio, spectral entropy, spectral roll-off point, spectral kurtosis, and Mel-frequency cepstral coefficients. The processing component 160 multiplies the target impact sound signal by a window function and performs a fast Fourier transform. The number of points in the Fast Fourier Transform can be 1024, 2048, 4096 or 8192, and the centroid of the spectrum is obtained by weighting the frequency and amplitude or power of each frequency point.

[0086] like Figures 1-7As shown, the main peak frequency is the frequency corresponding to the maximum power spectrum value within a preset frequency range. To avoid the influence of low-frequency mechanical vibration, the main peak search range can be set to 500Hz-20kHz. The frequency band energy ratio can be divided into frequency bands according to the actual acoustic spectrum of the impacting component 131 and the plastic particles. For example, it can be divided into four frequency bands: 0.5kHz-2kHz, 2kHz-5kHz, 5kHz-10kHz, and 10kHz-20kHz. The proportion of energy in each frequency band to the total energy is calculated. The spectral entropy is used to characterize the concentration of the spectral energy distribution. The Mel frequency cepstral coefficients can be extracted from the first 8 to 40 orders, preferably the 13th or 20th order. In addition to the static coefficients, the first-order and second-order difference coefficients can also be calculated. The attenuation characteristics can include the envelope attenuation coefficient, reverberation attenuation time, adjacent peak attenuation ratio, time for energy to drop to the preset proportion, and piecewise attenuation slope. The processing component 160 first performs full-wave rectification or Hilbert transform on the target impact acoustic signal to obtain the acoustic signal envelope, and then smooths the envelope. The envelope smoothing window can be 0.1ms-5ms. Starting from the peak moment of the impact envelope, a preset time period is selected for exponential decay fitting: A(t)=A0e^−λt; where A0 is the initial envelope amplitude, λ is the envelope decay coefficient, or the natural logarithm of the envelope can be taken for linear fitting, and the absolute value of the slope of the fitted line is used as the decay feature. The reverberation decay time can be defined as the time required for the envelope energy to decrease by 10dB, 20dB, 30dB or 60dB. Considering that the duration of the plastic particle impact signal is relatively short, the time required for the energy to decrease by 20dB or 30dB is used. The adjacent peak decay ratio can be defined as the ratio of the second envelope peak to the first envelope peak, or the average decay ratio between multiple consecutive local peaks. The time corresponding to the energy decreasing to the preset ratio can include the time required to decrease to 50%, 20%, 10%, 5% or 1% of the peak energy.

[0087] like Figures 1-7As shown, in order to improve the ability to identify changes in the proportion of damp and recycled materials and slight mixing, the time-frequency features, wavelet packet energy features, continuous wavelet transform features, short-time spectral texture features or deep acoustic embedding features of the acoustic signal can also be extracted. The processing component 160 can combine the above features into an acoustic feature vector: X = [x1, x2, ..., xn]. Before inputting the anomaly judgment model, each feature can be standardized. Standardization methods can include Z-score standardization, maximum-minimum normalization, or robust standardization based on median and interquartile range. Before formal testing, it is necessary to establish normal raw material acoustic signature benchmarks for different raw materials. These benchmarks include multiple normal acoustic signature samples of normal plastic granules under different feeding parameters and environmental conditions. When establishing benchmarks, normal plastic granules that have passed supplier inspection, laboratory inspection, or production verification should be selected. At least 100 valid impact events should be collected for each raw material, typically 500-5000. Raw material identification can include information such as material type, grade, supplier, batch, color, filler type, filler ratio, flame retardant rating, granule size, and allowable moisture content. For the same raw material, samples can be collected under different drop heights, different single feeding quantities, different ambient temperatures, and different ambient humidity levels. For example, drop heights can be set to 300mm, 400mm, and 500mm.

[0088] like Figures 1-7As shown, the single feeding quantity is set to 1, 3, and 5 pieces respectively; the ambient temperature covers 15℃-35℃; the relative humidity covers 30%-80%; the processing component 160 extracts the same voiceprint features from each normal voiceprint sample and determines the baseline value, allowable fluctuation range, and feature weight for each feature; the baseline value can be the arithmetic mean, median, or probability distribution center; the allowable fluctuation range can be the mean ± 2 standard deviations, the mean ± 3 standard deviations, percentile intervals, or confidence intervals; when the feature data is approximately normally distributed, the normal range can be set to μi ± 3σi; when the feature data is skewed, the 1st percentile to the 99th percentile can be used as the allowable fluctuation range; the feature weight can be... The ability to distinguish normal and abnormal samples based on different characteristics is determined. Weighting methods can include analysis of variance, Fisher's discriminant ratio, mutual information, random forest feature importance, logistic regression coefficient, or manual setting. The optimal weight for each feature is Σwi = 1. For moisture detection, the weights of envelope attenuation coefficient, spectral centroid, and high-frequency energy ratio can be increased; for incorrect material detection, the weights of main peak frequency, band energy ratio, and Mel frequency cepstral coefficient can be increased; for abnormal particle size, the weights of peak sound pressure, duration, and total energy can be increased. Normal raw material acoustic signatures can be stored in a database according to raw material identification. Database records should include at least the raw material identification, collection date, impactor type 131, and drop height. The acoustic signature reference should be recalibrated or updated after the impact component 131 is replaced, the drop height is adjusted, the sensor is replaced, or the main installation structure is changed. The processing component 160 compares the acoustic signature feature of the plastic particle 200 to be detected with the corresponding normal raw material acoustic signature reference to obtain the acoustic signature deviation degree. In a preferred calculation method, the standardized deviation of the i-th feature is calculated as follows: di = |xi - μi| / σi; where xi is the i-th acoustic signature feature of the event to be detected, μi is the reference value of the i-th feature of the normal sample, and σi is the standard deviation of the i-th feature of the normal sample. The acoustic signature deviation degree D can be calculated according to the following formula. D = Σwi·di; where wi is the weight of the i-th feature. To avoid excessive influence of a single abnormal feature on the result, an upper limit can be set for di, such as limiting di to the range of 0-10. In another implementation, the degree of voiceprint deviation is calculated using Mahalanobis distance to consider the correlation between each voiceprint feature: D = [(X-μ)ᵀS⁻¹(X-μ)]^1 / 2; where X is the voiceprint feature vector to be detected, μ is the mean vector of normal samples, and S is the covariance matrix of normal samples. When the number of samples is small or the covariance matrix is ​​close to singular, the covariance matrix can be regularized. The degree of voiceprint deviation can also be output by a single-class support vector machine, isolated forest, autoencoder, or Gaussian mixture model.

[0089] like Figures 1-7As shown, the preset anomaly judgment conditions may include one or more of the following: single deviation judgment, continuous event statistical judgment, and anomaly recognition model judgment. For single deviation judgment, when the voiceprint deviation degree D of a certain impact event is greater than a preset deviation threshold TD, the event is marked as an abnormal impact event. The preset deviation threshold can be determined based on the statistical distribution of the voiceprint deviation degree of normal samples; for example, the 99th percentile or 99.5th percentile of the deviation degree of normal samples can be set as the threshold. For continuous event statistical judgment, to reduce misjudgments caused by the randomness of single particle morphology, it is preferable to statistically analyze multiple consecutive impact events. For example, using 20 consecutive valid impact events as a detection window, when the voiceprint deviation degree is greater than the preset deviation threshold in more than 5 out of the 20 events, the event is considered an abnormal impact event. To determine if a batch of raw materials is abnormal, the detection window can include 10-200 valid impact events, preferably 20-100. The threshold for the number of abnormal events can be set to 10%-50% of the total number of events in the window. The processing component 160 can also calculate the percentage of abnormal events, the average deviation, and the dispersion of deviation. The percentage of abnormal events is the ratio of the number of abnormal impact events to the total number of valid impact events. The average deviation is the average or median of the sound signature deviation of each impact event within the window. The dispersion of deviation can be expressed using standard deviation, coefficient of variation, range, or interquartile range. For material mismatch anomalies, most impact events are usually significantly inconsistent with the target baseline, resulting in a high percentage of abnormal events and a large average deviation. For small amounts of mixed materials... Normally, only a portion of impact events show significant deviations, resulting in a low to moderate proportion of anomalous events and a large dispersion in the degree of deviation. For dampness anomalies, most events may exhibit moderate deviations in both attenuation and high-frequency energy characteristics. For impurity contamination anomalies, a few impact events with extremely large deviations may occur, accompanied by anomalies in peak values, main peak frequencies, or high-frequency energy. The anomaly recognition model processing component 160 can call a pre-trained anomaly recognition model. The input to the anomaly recognition model is the voiceprint feature vector, the degree of voiceprint deviation, and the statistical features of continuous events. The output is the anomaly probability and the anomaly type probability. The anomaly recognition model can employ support vector machines, random forests, gradient boosting decision trees, artificial neural networks, or convolutional neural networks. For recurrent neural networks or Transformer models, random forests, support vector machines, or gradient boosting decision trees are preferred for feature vector inputs. For original waveforms or time-frequency graphs, one-dimensional or two-dimensional convolutional neural networks can be used. The model training data includes normal samples and known abnormal samples. For each raw material and each type of abnormality, at least 500 valid collision samples should be collected. The training data is divided into training, validation, and test sets, for example, in a ratio of 70%, 15%, and 15%. Different batches of raw materials are preferably assigned to different datasets to verify the model's adaptability to batch variations. When the abnormal probability output by the model is greater than a preset probability threshold, it is judged as an abnormality. The probability threshold can be 0.5-0.99, preferably 0.7-0.95.

[0090] like Figures 1-7 As shown, the processing component 160 acquires the raw material identifier corresponding to the plastic granules 200 to be detected and calls the normal raw material acoustic signature corresponding to the raw material identifier. The raw material identifier can be input by the operator, or it can be obtained automatically by scanning the barcode, QR code, RFID tag on the raw material packaging or from the production execution system. The processing component 160 determines the anomaly type and anomaly level based on the proportion of abnormal events corresponding to continuous impact events, the average deviation, the dispersion of the deviation, and the output probability of various models. An example judgment logic is as follows: when the proportion of abnormal events is greater than 80%, the average deviation is large, and the acoustic signature features are generally close to the acoustic signature of another registered raw material, it is judged as a material error anomaly. When the proportion of abnormal events is 5%-60%, the dispersion of the deviation is large, and the abnormal events can be clustered into the acoustic signature of another material, it is judged as a mixed material anomaly. When the high-frequency energy ratio of most events decreases, the spectral centroid drops, or the duration or decay time changes, it is judged as a mixed material anomaly. When the change is stable and the direction of change is consistent with that of the damp sample, it is judged as a dampness abnormality. When a small number of events show a significant increase in peak sound pressure, a sudden change in the main peak frequency, or an enhancement of high-frequency impact components, it is judged as an impurity contamination abnormality. When the overall sound signature is between that of virgin material and recycled material, and the deviation corresponds to the recycled material ratio calibration curve, it is judged as a recycled material ratio abnormality. When the peak energy, duration, and energy of some frequency bands show a multi-peak distribution and correspond to calibration samples of different particle sizes, it is judged as a particle size abnormality. The abnormality level can be divided into four levels: warning, general, serious, and emergency, or into three levels: level one, level two, and level three. For example, when the proportion of abnormal events is less than 10% and the average deviation is slightly higher than the threshold, it is judged as a warning level; when the proportion of abnormal events is 10%-30%, it is judged as a general level; when the proportion of abnormal events is 30%-70%, it is judged as a serious level; when the proportion of abnormal events is greater than 70% or it is identified as incorrect material, it is judged as an emergency level.

[0091] like Figures 1-7As shown, after the abnormal result output and linkage control processing component 160 generates the raw material abnormality detection result, it can output the result through a display screen, audible and visual alarm, industrial Ethernet, RS485, CAN bus, or digital output interface. The abnormal detection result can include raw material identification, detection time, number of valid impact events, number of superimposed impact events, abnormal event ratio, average deviation, abnormality type, abnormality level, and handling suggestions. When configuring the abnormal raw material handling component, the abnormal raw material handling component can include a flap valve, a rotary distributing valve, a pneumatic push rod, or an air blowing sorting mechanism. When the detection result is normal, the processing component 160 controls the distributing mechanism to make the particles enter the qualified material channel; when the detection result is abnormal, it controls the distributing mechanism to switch to the abnormal material channel and can stop the subsequent feeding equipment. The distributing mechanism should be set downstream of the impactor 131 and reserve a particle conveying delay corresponding to the detection and processing time. The processing component 160 can delay the output of the distributing control command according to the time required for the particles to move from the impact area to the distributing position.

[0092] like Figures 1-7As shown, after the device is started, the processing component 160 reads the current raw material identification and detection formula. The detection formula includes at least the type of impactor 131, drop height, single feeding quantity, feeding interval, vibration trigger threshold, effective impact time duration, filtering parameters, acoustic signature reference, and anomaly judgment threshold. The processing component 160 checks the communication status of each sensor and collects background sound and background vibration signals for 5-30 seconds under no-particle feeding conditions to update the background noise level and dynamic trigger threshold. Subsequently, the processing component 160 sends a feeding command to the discharge control mechanism 113. The quantitative discharge mechanism 112 outputs a single or preset number of plastic particles. The plastic particles enter the dropping channel 121, and after being restricted lateral movement by the attitude rectification structure 122, they fall freely from the discharge end 124 of the dropping channel 121. The plastic particles impact the preset impact area 137 of the impactor 131 at the preset drop height. The impactor 131 generates mechanical vibration after being impacted, and Impact sound radiates into the surrounding air. The main sound sensor 142 collects the impact sound signal, the vibration sensor 144 collects the impact vibration signal, and the reference sound sensor 151 simultaneously collects the background interference signal. The processing component 160 determines the impact trigger time and effective impact period based on the vibration signal, and extracts the effective impact sound signal from the continuous sound signal. The processing component 160 uses the background interference signal to suppress the interference of the effective impact sound signal to obtain the target impact sound signal. The processing component 160 extracts the time domain characteristics, frequency domain characteristics, and attenuation characteristics of the target impact sound signal and compares them with the corresponding normal raw material acoustic signature. The degree of acoustic signature deviation is calculated. The processing component 160 performs statistics on multiple consecutive impact events and generates an anomaly detection result when the degree of acoustic signature deviation, the number of abnormal events, or the output of the anomaly recognition model meets the preset anomaly judgment conditions. After completing one detection window, the processing component 160 can continue to perform the next window detection to achieve continuous or periodic detection.

[0093] Example 1: Detection of mismatched and mixed polypropylene granules

[0094] This embodiment is used to detect whether polyethylene particles are mixed in or polyethylene raw materials are mistakenly used when the target raw material is polypropylene granules. The storage silo 111 is made of 304 stainless steel with an effective volume of 5L. The bottom discharge port 114 has a diameter of 20mm. The quantitative discharge mechanism 112 uses an 18-hole rotating quantitative disk, which outputs one plastic granule at a time. A stepper motor drives the quantitative disk to rotate 20° every 350ms. The discharge channel 121 is a transparent polycarbonate tube with an inner diameter of 12mm, a wall thickness of 2mm, and a length of 420mm. The vertical distance between the bottom of the discharge channel 121 and the surface of the impactor 131 is 400mm. The bottom of the feeding channel 121 is equipped with a polyoxymethylene limiting nozzle with an outlet inner diameter of 7mm. The impact component 131 is made of 304 stainless steel round plate with a diameter of 180mm and a thickness of 1.5mm. The edge of the impact component 131 is clamped by an annular support seat and an upper pressure ring with a clamping width of 10mm. Four solid silicone rubber vibration isolation blocks with a diameter of 25mm, a height of 20mm, and a Shore A hardness of 50 are evenly arranged below the annular support seat. The main sound sensor 142 is a digital MEMS microphone with a frequency response range of 20Hz-20kHz, which is set 70mm below the impact component 131. Vibration sensor 1 A triaxial MEMS accelerometer with a range of ±32g and a bandwidth of 5kHz is used, bonded to the lower surface of the impactor 131 at a distance of 25mm from the edge. A reference acoustic sensor 151 is positioned outside the impact detection assembly, 600mm from the center of the impact area, and separated from the impactor 131 by a metal sound insulation plate. The acoustic signal sampling frequency is 48kHz, and the quantization accuracy is 24 bits. The vibration signal sampling frequency is 24kHz. The processing component 160 performs a 200Hz-8000Hz bandpass filter on the vibration signal and obtains the vibration envelope using a 1ms sliding root mean square. (Background label...) The time interval was set to 10 seconds, and the trigger threshold was set to the average background envelope plus 6 standard deviations. The forward time of the effective impact period was 10 ms, and the backward time was 120 ms. After adaptive noise cancellation and bandpass filtering from 500 Hz to 18 kHz, the peak sound pressure level, root mean square value, kurtosis, duration, spectral centroid, main peak frequency, energy ratio of the four frequency bands, spectral entropy, 13th order Mel frequency cepstral coefficient, envelope attenuation coefficient, and time required for a 20 dB energy decrease were extracted. 2000 effective impact events were collected using normal polypropylene particles to establish a normal raw material acoustic signature benchmark. During detection, 50 consecutive effective impact events were used as a window. When the deviation of a single acoustic signature was greater than 3.5, the event was marked as abnormal. When the number of abnormal events reached 8 out of 50 events, a mixing abnormality was determined. When the number of abnormal events reached more than 40, and the distance between the acoustic signature and the polyethylene benchmark was less than the distance between the acoustic signature and the polypropylene benchmark, a material error abnormality was determined.

[0095] Example 2: Moisture Detection of Polyamide Particles

[0096] This embodiment is used to detect abnormal moisture absorption in nylon plastic granules. The raw material to be tested is PA66 granules, with a normal moisture content requirement of no more than 0.2%. The impactor 131 is a stainless steel plate with a diameter of 160 mm and a thickness of 1.2 mm. The drop height is 500 mm, the single feed quantity is 3 granules, and the feed interval is 500 ms. When establishing the benchmark, PA66 samples with moisture contents of 0.05%, 0.1%, 0.2%, 0.4%, 0.8%, and 1.2% were prepared respectively. At least 1000 effective impact events were collected for each moisture content sample, with a focus on extracting high-frequency energy. The device calculates the ratio, spectral centroid, peak sound pressure level, envelope attenuation coefficient, and time required for a 20dB energy drop, and establishes a regression relationship between moisture content and acoustic signature characteristics. The processing component 160 uses a gradient boosting regression model to output an estimated moisture content. When the estimated moisture content is greater than 0.2%, and more than 20 out of 30 consecutive detection events exceed the corresponding threshold, a dampness anomaly result is generated. To reduce the impact of changes in ambient humidity, a temperature and humidity sensor is installed inside the device. The processing component 160 compensates for the acoustic signature reference based on ambient temperature and relative humidity. Temperature compensation can be achieved through linear compensation or lookup table compensation.

[0097] Before using the device, check whether there are any plastic granules left in the storage bin 111, the quantitative discharge mechanism 112, and the discharge channel 121. When changing the type of raw material, clean the parts that come into contact with the granules. The surface of the impact component 131 should be kept clean. If there are obvious scratches, pits, attachments, or corrosion on the surface of the impact component 131, the impact component 131 should be replaced and recalibrated. The main sound sensor 142 can be calibrated for sensitivity using a standard sound calibrator. The vibration sensor 144 can be calibrated using a standard vibration table or a known impact source. The device can perform a self-test when it is turned on every day. During the self-test, the electromagnetic impactor strikes the impact component 131 with a fixed energy. The processing component 160 compares the sound of this self-test with the equipment self-test benchmark. When the difference exceeds the allowable range, it prompts to check the impact component 131, the sensor, or the installation structure. In order to compensate for long-term drift, the normal sound benchmark can be updated periodically using standard normal raw materials. When updating, the original benchmark and update record should be retained to prevent abnormal raw materials from being mistakenly included in the normal benchmark.

[0098] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A raw material anomaly detection device based on the impact sound signature of plastic particles, characterized in that, include: Feeding assembly: used to receive the plastic particles (200) to be tested and output the plastic particles (200) to be tested according to preset feeding parameters; Material dropping constraint component: disposed on the discharge side of the feeding component, used to limit the falling path of the plastic particle to be tested (200) so that the plastic particle to be tested (200) is output according to the preset impact conditions; Impact detection assembly: includes an impact member (131) and a support structure (133) for supporting the impact member (131). The impact member (131) is disposed below the material discharge restraint assembly and is used to receive the plastic particles (200) to be tested output by the material discharge restraint assembly and generate an impact response when the plastic particles (200) to be tested hit the impact member (131). Signal acquisition component: includes an acoustic signal acquisition unit and a vibration signal acquisition unit. The acoustic signal acquisition unit is used to acquire the impact acoustic signal generated when the plastic particle (200) to be detected impacts the impactor (131). The vibration signal acquisition unit is connected to the impact detection component to acquire the impact vibration signal generated by the impactor (131). Interference monitoring component: used to collect background interference signals in the environment where the impact detection component is located; Processing component (160): connected to the feeding component, signal acquisition component, and interference monitoring component respectively, the processing component (160) is configured to: determine the effective impact period based on the impact vibration signal; extract the corresponding effective impact sound signal from the impact sound signal according to the effective impact period; perform interference suppression processing on the effective impact sound signal based on the background interference signal to obtain the target impact sound signal; extract impact acoustic features from the target impact sound signal, including at least time domain features, frequency domain features, and attenuation features; and compare the impact acoustic features with a pre-established normal raw material acoustic feature benchmark to obtain the acoustic deviation degree of the plastic particle (200) to be detected. When the degree of deviation of the voiceprint meets the preset abnormality judgment condition, the raw material abnormality detection result is generated.

2. The raw material anomaly detection device based on the impact sound signature of plastic particles according to claim 1, characterized in that, The feeding assembly includes a storage bin (111), a quantitative discharging mechanism (112), and a discharging control mechanism (113). The quantitative discharging mechanism (112) is located at the discharge port (114) of the storage bin (111) and is used to output a preset quantity or preset quality of plastic particles (200) to be tested at preset time intervals. The discharging control mechanism (113) is connected to the processing assembly (160) and is used to adjust the discharging frequency and discharging amount of the quantitative discharging mechanism (112) according to the feeding control command output by the processing assembly (160) so that two adjacent particle collision events are separated from each other in time.

3. The raw material anomaly detection device based on the impact sound of plastic particles as described in claim 1, characterized in that, The material dropping constraint assembly includes a vertically extending material dropping channel (121) and an attitude rectification structure (122) disposed within the material dropping channel (121). The inlet end (123) of the material dropping channel (121) is connected to the outlet side of the feeding assembly, and the outlet end (124) of the material dropping channel (121) faces the impact member (131). The attitude rectification structure (122) is used to reduce the lateral motion component of the plastic particle (200) to be tested and to assist the plastic particle (200) to be tested to fall from a preset height range to the preset impact area (137) of the impact member (131).

4. The raw material anomaly detection device based on the impact sound signature of plastic particles according to claim 1, characterized in that, The impactor (131) is detachably mounted on the support structure (133). The impactor (131) has an impact surface for impact by the plastic particles (200) to be tested. A vibration isolation member (134) is provided between the support structure (133) and the mounting base (135). The vibration isolation member (134) is used to suppress the transmission of mechanical vibration of the mounting base (135) to the impactor (131). At least one of the material, thickness and effective impact area of ​​the impactor (131) is preset according to the raw material type of the plastic particles (200) to be tested.

5. The raw material anomaly detection device based on the impact sound signature of plastic particles according to claim 1, characterized in that, The acoustic signal acquisition unit includes at least one main acoustic sensor (142), which is disposed to the side or below the impactor (131) and spaced apart from the impactor (131). The vibration signal acquisition unit includes a vibration sensor (144) fixed on the impactor (131) or the support structure (133). The interference monitoring component includes a reference acoustic sensor (151), which is disposed outside the impact detection component and away from the preset impact area (137) so that the environmental noise component in the background interference signal acquired by the reference acoustic sensor (151) is greater than the impact sound component.

6. The raw material anomaly detection device based on the impact sound signature of plastic particles according to claim 1, characterized in that, The processing component (160) is configured to: filter the impact vibration signal to obtain a vibration envelope signal; The impact trigger time is determined when the vibration envelope signal exceeds a preset vibration trigger threshold. Based on the impact trigger time, an effective impact period is constructed according to a preset forward time length and a preset backward time length; When two adjacent valid impact periods overlap, the corresponding impact events are marked as superimposed impact events, and the valid impact sound signals corresponding to the superimposed impact events are removed or the valid impact sound signals corresponding to the superimposed impact events are processed by sound source separation.

7. The raw material anomaly detection device based on the impact sound signature of plastic particles according to claim 1, characterized in that, The time-domain features include at least two of the following: peak sound pressure level, root mean square value, waveform factor, impulse factor, kurtosis, and duration. The frequency-domain features include at least two of the following: spectral centroid, main peak frequency, band energy ratio, spectral entropy, and Mel frequency cepstral coefficient. The attenuation features include at least one of the following: envelope attenuation coefficient, reverberation attenuation time, adjacent peak attenuation ratio, and the time corresponding to the energy decreasing to a preset ratio.

8. The raw material anomaly detection device based on the impact sound signature of plastic particles according to claim 1, characterized in that, The normal raw material acoustic signature includes multiple normal acoustic signature samples corresponding to normal plastic particles under different feeding parameters and environmental conditions. The processing component (160) is configured to determine the reference value, allowable fluctuation range and feature weight of each impact acoustic signature feature based on the multiple normal acoustic signature samples, and calculate the acoustic signature deviation degree based on the deviation of the impact acoustic signature feature relative to the reference value and the feature weight. The preset anomaly judgment condition includes at least one of the following conditions: the acoustic signature deviation degree is greater than the preset deviation threshold. The number of impact events in which the voiceprint deviation exceeds a preset deviation threshold within a preset number of consecutive impact events reaches a preset number; The anomaly probability output by the pre-trained anomaly recognition model is greater than a preset probability threshold.

9. A method for detecting raw material anomalies based on the impact sound of plastic particles, comprising the raw material anomaly detection device based on the impact sound of plastic particles as described in any one of claims 1-8, characterized in that, include: The plastic particles (200) to be tested are output according to the preset feeding parameters and the falling path of the plastic particles (200) to be tested is limited so that the plastic particles (200) to be tested impact the impactor (131) according to the preset impact conditions. The impact sound signal and impact vibration signal generated when the plastic particle (200) to be tested impacts the impactor (131) are collected, and the background interference signal in the environment where the impactor (131) is located is collected. The effective impact period is determined based on the impact vibration signal; The corresponding valid impact sound signal is extracted from the impact sound signal according to the effective impact time period; Based on the background interference signal, the effective impact sound signal is subjected to interference suppression processing to obtain the target impact sound signal; Extract impact acoustic signature features from the target impact acoustic signal, including at least time-domain features, frequency-domain features, and attenuation features; The impact acoustic signature is compared with a pre-established acoustic signature benchmark for normal raw materials to obtain the degree of acoustic signature deviation of the plastic particle (200) to be tested; When the degree of deviation of the voiceprint meets the preset abnormality judgment condition, the raw material abnormality detection result is generated.

10. The raw material anomaly detection method based on the impact sound signature of plastic particles according to claim 9, characterized in that, The method further includes: obtaining the raw material identifier corresponding to the plastic particles (200) to be detected and calling the corresponding normal raw material acoustic signature according to the raw material identifier; The corresponding voiceprint deviation degree is obtained based on a preset number of consecutive impact events; The deviation of voiceprints corresponding to the preset number of consecutive impact events is statistically analyzed to obtain at least one of the following: the proportion of abnormal events, the average deviation, and the dispersion of deviation. The abnormality type and abnormality level of the plastic granules (200) to be tested are determined based on at least one of the abnormal event proportion, average deviation degree, and deviation degree dispersion; wherein the abnormality type includes at least one of the following: mixing abnormality, wrong material abnormality, moisture abnormality, impurity mixing abnormality, recycled material ratio abnormality, and granule size abnormality.

Images