Multimodal Defect Detection System and Multidimensional Detection Method for Capsule Drugs

Abstract

The present invention relates to the technical field of pharmaceutical quality inspection, specifically a multi-modal defect detection system and multi-dimensional detection method for capsule drugs. The detection system includes a detection container filled with an intelligent-responsive pharmaceutical-grade silicone oil composition, a transmission system, a detection mechanism, a control system, and a rejection mechanism. Through density adjustment, temperature-sensitive-fluorescent response, and multi-modal detection technologies, precise detection of defects such as capsule liquid leakage, cracks, and poor sealing can be achieved. The detection method adjusts the silicone oil density according to the capsule type, and collects and fuses and analyzes buoyancy signals and optical image data to judge defects. The invention has high detection accuracy, can detect liquid leakage at the level of 0.02 μL and cracks at the level of 0.1 mm; has strong compatibility, is applicable to more than 95% of commercially available capsules; has low cost, with significantly reduced equipment and maintenance costs, and the silicone oil can be recycled; and complies with pharmaceutical production specifications.

Description

Technical Field

[0001] The present invention relates to the technical field of pharmaceutical quality inspection, specifically a multi-modal defect detection system and multi-dimensional detection method for capsule drugs. Background Art

[0002] In the pharmaceutical industry, capsule drugs are widely used due to their advantages such as easy administration and accurate dosage. However, during the production process of capsules, due to various factors such as die wear, uneven filling pressure, and differences in raw material quality, defects such as liquid leakage, cracks, and poor sealing are likely to occur. These defects not only affect the quality and efficacy of drugs but may also pose a potential threat to the health of patients.

[0003] Traditional capsule defect detection methods mainly include manual visual inspection and simple physical detection means. Manual visual inspection relies on the experience and eyesight of inspectors, with low detection efficiency, poor accuracy, and it is easy for inspectors to get fatigued during long-term work, resulting in a relatively high missed detection rate. For example, it is very difficult to detect tiny cracks or liquid leakage manually. Simple physical detection means, such as buoyancy detection, although it can detect some liquid leakage problems, has low detection sensitivity for tiny liquid leakage. Usually, it can only detect liquid leakage above 0.1 μL, and the detection success rate for leak holes below 0.05 μL is less than 70%. Moreover, this method is easily interfered by bubbles and impurities, leading to misjudgment.

[0004] In terms of optical detection, the resolution of existing ordinary optical imaging devices is limited, and it is difficult to capture fine cracks below 0.1 mm on the surface of capsules. At the same time, these methods cannot effectively detect defects inside the capsules. In addition, existing detection devices often can only adopt a single detection principle and cannot comprehensively judge based on multiple detection information, resulting in low reliability of detection results.

[0005] In terms of detecting liquids, although commonly used pharmaceutical-grade dimethyl silicone oil has certain inertness, it has many deficiencies. Its single density characteristic is only applicable to gelatin capsules (density 1.01 g / mL). For other types of capsules such as HPMC plant capsules (density 1.03 g / mL), detection liquids need to be configured separately, which greatly reduces the compatibility of the equipment, increases the detection cost and the complexity of operation. Moreover, during long-term recycling, nano-particles in the silicone oil are prone to deposition, affecting the detection accuracy. Usually, after 500 uses, the detection efficiency will decrease by 15%, and the liquid needs to be replaced frequently, further increasing the production cost and time cost. The detection success rate of traditional buoyancy detection for leak holes below 0.05 μL is less than 70%, and the misjudgment rate is as high as 15%; the missed detection rate of ordinary optical detection for cracks below 0.1 mm level exceeds 30%, and single-density silicone oil is only applicable to gelatin capsules, and separate liquid preparation is required for HPMC plant capsules, increasing the detection cost by more than 40%.

[0006] With the continuous improvement of the pharmaceutical industry's requirements for drug quality, it is urgent to develop a capsule drug defect detection system and method with high detection accuracy, strong compatibility, and the ability to achieve multi-dimensional detection. Summary of the Invention

[0007] The present invention aims to provide a multi-modal defect detection system and a multi-dimensional detection method for capsule drugs to solve the problems existing in the existing detection technologies, such as low detection sensitivity, poor equipment compatibility, single detection means, and short service life of liquids, and to achieve efficient, accurate, and multi-dimensional detection of capsule drug defects.

[0008] The technical solution adopted by the present invention to solve its technical problems is: a multi-modal defect detection system and a multi-dimensional detection method for capsule drugs, including: a detection container filled with an intelligent response type pharmaceutical grade silicone oil composition, the composition containing by mass percentage: 86-92% of pharmaceutical grade dimethyl silicone oil (50cSt, density 0.96g / mL), 3-6% of hydroxylated silica nanospheres (particle size 50nm), 1-2% of thermosensitive polyethylene glycol grafted silicone oil (PEG molecular weight 2000Da), 0.1-0.5% of fluorescein sodium derivative, and the balance being inevitable impurities, and the sum of the mass percentages of each component is 100%; a transmission system, including a horizontally arranged mesh belt conveyor, the conveyor belt being immersed 5-10mm below the silicone oil liquid surface and moving uniformly at a speed of 0.2-2m / s to carry capsules; a detection mechanism, including a pressure sensor array arranged at the bottom of the detection container and optical imaging units symmetrically arranged on both sides of the conveyor belt; a control system, electrically connected to the pressure sensor array and the optical imaging units, and built-in with a machine learning algorithm to fuse detection data to judge capsule defects; a rejection mechanism, arranged at the end of the conveyor belt to reject defective capsules according to the instructions of the control system. This system realizes high-sensitivity, high-compatibility, and low false alarm rate capsule defect detection through the intelligent response type silicone oil composition, multi-modal detection technology, and intelligent algorithm fusion.

[0009] Specifically, the thermosensitive polyethylene glycol grafted silicone oil is in a stretched state under a 30°C detection environment. When leakage causes a local temperature increase (≥5°C), the PEG chain segment dehydrates and contracts, causing the density of the leakage area to suddenly increase by 0.05g / mL, forming a characteristic buoyancy pulse that can be captured by the pressure sensor (response time <0.1 second); the fluorescein sodium derivative only specifically reacts with hydroxyproline, the degradation product of gelatin, and the fluorescence intensity increases by more than 10 times after the reaction. Combining with the 700nm band-pass filter of the high-speed camera, the fluorescence enhancement signal of a 0.02μL-level leakage point can be identified.

[0010] Specifically, the surface tension of the pharmaceutical-grade dimethyl silicone oil is 20-25 mN / m, the moisture content is <0.1%, and it is circularly filtered through a 0.22-μm sterilizing filter to ensure the cleanliness of the silicone oil, reduce the interference of impurities on the detection, and effectively improve the stability and accuracy of the detection.

[0011] Specifically, the pharmaceutical-grade dimethyl silicone oil is circularly filtered through a 0.22-μm sterilizing filter, and the total number of colonies is ≤10 CFU / mL; the triclosan loading is 0.01%, meeting the 0.3% safety threshold specified in USP<51> and FDA regulations.

[0012] Specifically, the pressure sensor array includes at least 10 distributed piezoresistive sensors with an accuracy of ±0.01 kPa and a sampling frequency of ≥100 Hz, which can detect the buoyancy change caused by a liquid leakage of 0.05 μL level, significantly improving the sensitivity of liquid leakage detection.

[0013] Specifically, the red laser of the optical imaging unit has a wavelength of 650 nm, the frame rate of the high-speed camera is ≥500 fps, and the resolution is ≥1280×1024. It can capture cracks on the capsule surface at the 0.1-mm level, has a higher resolution, and can detect finer surface defects.

[0014] Specifically, the machine learning algorithm of the control system is a support vector machine (SVM) model. The pre-trained data includes at least 100,000 buoyancy-image feature pairs of normal / defective capsules, and the defect recognition accuracy is ≥99.9%, effectively reducing the false alarm rate, which is essentially different from the traditional method relying on manual experience or simple threshold judgment.

[0015] Specifically, the control system extracts the time-domain features (fluctuation amplitude, frequency), frequency-domain features (energy spectral density) of the buoyancy signal, as well as the edge curvature, gray value distribution, and fluorescence intensity threshold of the optical image, and matches the pre-trained 100,000+ buoyancy-image feature pairs through the support vector machine (SVM) model to achieve the classification of three types of defects: liquid leakage, cracks, and normal, with a false alarm rate of ≤0.05%.

[0016] Specifically, the inner wall of the detection container is coated with a polytetrafluoroethylene coating with a surface roughness Ra ≤ 0.2 μm, and the conveyor belt is made of 316L stainless steel with a mesh hole diameter of 1 / 2-2 / 3 of the capsule diameter, optimizing the equipment structure to meet the detection requirements and reducing the interference on capsule detection.

[0017] Specifically, the density of the intelligent response type pharmaceutical grade silicone oil composition can be dynamically adjusted within the range of 0.98 - 1.02 g / mL by the concentration of silica nanospheres, the surface tension is 20 - 25 mN / m, and the fluorescence intensity increases by more than 10 times when contacting the leakage liquid containing hydroxyproline, enhancing the detection ability of the leakage liquid, breaking through the limitation of the traditional detection of a single density of the liquid, and realizing the compatible detection of various types of capsules.

[0018] Specifically, a microfluidic diversion groove (depth 0.5 mm, spacing 5 mm) is provided on the inner wall of the detection container to guide the intelligent response type pharmaceutical grade silicone oil composition to form a stable laminar flow, prevent the deposition of nanoparticles, and ensure the stability of the detection.

[0019] Specifically, the high-speed camera of the optical imaging unit is integrated with a 700 nm band-pass filter to collect the specific fluorescence images of fluorescein sodium derivatives, assist in defect judgment, and provide a fluorescence detection dimension.

[0020] A multi-dimensional detection method for capsule drugs, including: adjusting the density of the intelligent response type pharmaceutical grade silicone oil composition according to the capsule type; the pressure sensor array collects the buoyancy fluctuation signal, and the optical imaging unit collects the optical image; the control system fuses and analyzes the data, and judges whether the capsule has leakage, cracks or surface defects according to the preset threshold; if it is determined to be a defective capsule, the rejection mechanism will reject it. This method integrates multi-dimensional detection data and realizes more accurate and efficient detection of capsule defects.

[0021] The beneficial effects of the present invention:

[0022] The detection accuracy is significantly improved: Through the temperature-sensitive - fluorescence dual-response technology and multi-modal data fusion, accurate identification of leakage liquid at the 0.02 μL level (the lower detection limit of the traditional method is 0.1 μL) and cracks at the 0.1 mm level is realized, the defect identification accuracy rate is ≥99.9%, the false alarm rate is as low as 0.01%, the sensitivity is increased by 20 times compared with the traditional single detection method, and the reliability is increased by more than 3 times.

[0023] Compatibility with all types of capsules: Based on the density dynamic adjustment technology of silica nanospheres (0.98 - 1.02 g / mL), a single composition can be adapted to more than 95% of commercially available capsules such as gelatin (1.01 g / mL), HPMC plant capsules (1.03 g / mL), pullulan polysaccharide capsules (1.00 g / mL), etc. There is no need to separately configure detection liquids for different capsules, the versatility of the equipment is increased by 3 times, and the degree of automation of the detection process is significantly improved.

[0024] Outstanding cost and efficiency advantages: The equipment cost is only 1 / 5 of that of X-ray detection systems. The service life of silicone oil circulation is extended to 1000 times (500 times by traditional methods), the maintenance cost is reduced by 70%, and the comprehensive detection cost is decreased by 30%. The single-machine production capacity can reach 1200 capsules per minute (taking gelatin capsules as an example), significantly improving the detection efficiency of the production line.

[0025] Compliance and safety guarantee: Pharmaceutical-grade dimethyl silicone oil certified by ICH Q7 is adopted, and all components comply with the pharmacopoeia standards of USP, EP, and ChP. The triclosan loading is ≤ 0.01% (lower than the 0.3% safety threshold stipulated by the FDA). The detection containers pass the CIP / SIP cleaning validation, the silicone oil pollution degree (TOC) < 5 ppm, and the total number of colonies ≤ 10 CFU / mL, fully meeting the compliance requirements of GMP and FDA 21 CFR Part 11 for pharmaceutical production detection equipment.

[0026] Multi-dimensional defect determination ability: By integrating the time-domain / frequency-domain characteristics of buoyancy signals, the edge curvature / gray-scale distribution of optical images, and the fluorescence intensity threshold, a three-level defect determination model (leakage / crack / normal) is constructed, which can effectively distinguish various defect types such as leakage, cracks, and poor sealing, avoiding the risk of missed judgment of traditional single detection principles, and increasing the detection coverage by 40%. Description of the Drawings

[0027] The present invention will be further described below in conjunction with the drawings and embodiments.

[0028] Figure 1 is the system architecture diagram of the present invention;

[0029] Figure 2 is the method flow diagram of the present invention. Detailed Embodiments

[0030] In order to make the technical means, creative features, achieved purposes, and effects of the present invention easy to understand, the present invention will be further described below in conjunction with specific embodiments.

[0031] As Figure 1 - Figure 2 shown, the present invention provides the following technical solutions:

[0032] Detection system:

[0033] Detection Container: It is made of 316L stainless steel with a polytetrafluoroethylene coating on the inner wall to reduce liquid adhesion. It is filled with an intelligent responsive pharmaceutical-grade silicone oil composition, which contains a variety of functional components. Among them, pharmaceutical-grade dimethyl silicone oil is used as the basic component, which has passed the ICH Q7 certification and has no chemical reaction with the capsules. After soaking for 72 hours, the weight change of the capsules is <0.01%; hydroxylated silica nanospheres are used to adjust the density of the silicone oil, making it continuously adjustable within the range of 0.98 - 1.02 g / mL to adapt to different types of capsules; thermosensitive polyethylene glycol-grafted silicone oil can dynamically mark liquid leakage; sodium fluorescein derivatives are used to specifically identify liquid leakage. Compared with traditional detection liquids, this composition realizes the dynamic adjustment of density and the specific detection of liquid leakage, solving the problems of the limited single density of traditional detection liquids and the detection blind area of tiny liquid leakage.

[0034] Drive System: The mesh belt conveyor is woven with food-grade stainless steel wires, and the mesh hole diameter matches the capsule size (for example, for 00# capsules, the corresponding mesh hole is 3 mm), ensuring that the capsules float freely on the conveyor belt and avoiding mechanical contact interference with the detection. The conveyor belt speed can be adjusted within the range of 0.2 - 2 m / s to meet the detection requirements of different specifications of capsules.

[0035] Detection Mechanism: The pressure sensor array is distributed at the bottom of the detection container to collect the pressure fluctuation signals in real time when the capsules are floating. The detection lower limit can reach 0.05 μL of liquid leakage. The optical imaging unit consists of a red light laser and a high-speed camera symmetrically arranged on both sides of the conveyor belt. The red light laser projects parallel light onto the surface of the capsules, and the high-speed camera collects 360° images. Combining with the background subtraction algorithm, it can identify defects such as cracks and depressions with a size of 0.1 mm on the surface of the capsules. Compared with traditional detection methods, this detection mechanism has significantly improved sensitivity and resolution in liquid leakage and crack detection.

[0036] Control System: It adopts an industrial-grade PLC + host computer architecture and incorporates a support vector machine (SVM) algorithm model. This model integrates the buoyancy signals collected by the pressure sensors (time-domain features: fluctuation amplitude, frequency; frequency-domain feature: energy spectral density) and the optical image features collected by the optical imaging unit (edge curvature, gray value distribution). Through the threshold adaptive algorithm (the training data contains more than 100,000 defect samples), it can accurately classify the capsule defects (liquid leakage, cracks, normal), and the false alarm rate is ≤0.05%. Different from the traditional method that relies on manual experience or simple threshold judgment, this control system uses machine learning algorithms to achieve intelligent and high-precision defect judgment.

[0037] Rejection Mechanism: It uses a pneumatic nozzle with a response time <50 ms. Within 0.1 second after detecting the defect signal, it sprays a high-pressure air flow of 0.4 - 0.6 MPa to blow the defective capsules off the conveyor belt and into the waste collection box.

[0038] Detection Method:

[0039] First, according to the type of the capsule to be detected (such as gelatin, HPMC plant capsule, etc.), the control system realizes automatic adjustment by controlling the feeding pump to pump silica nanospheres, adjusts the density of the intelligent responsive pharmaceutical-grade silicone oil composition, and enables the normal capsule to present a suitable floating state in the silicone oil (for example, the gelatin capsule is in a semi-floating state in the silicone oil with a density of 0.98 g / mL, and the immersion depth is 2 / 3; the HPMC capsule is in a semi-floating state in the silicone oil with a density of 1.02 g / mL, and the immersion depth is 70%).

[0040] Then, the pressure sensor array collects the buoyancy fluctuation signals of the capsule in real time, and the optical imaging unit synchronously collects the optical images of the capsule surface. The control system performs fusion analysis on the collected buoyancy signals and optical image data, and extracts relevant features. For the leaking capsule, the standard deviation of the buoyancy signal will be greater than that of the normal capsule (such as the standard deviation of the buoyancy signal of the leaking capsule > 0.1 kPa, and that of the normal capsule ≤ ±0.05 kPa), and liquid leakage traces may be shown in the optical image.

[0041] Finally, according to the preset thresholds determined by statistical analysis of historical experimental data (such as the pressure sensor signal fluctuation threshold is ±0.01 kPa, and the optical image gray value threshold is 120) and the classification model constructed by using the support vector machine (SVM) algorithm and trained with 100,000 groups of labeled samples (including categories such as leaking, cracked, qualified, etc.), it is judged whether the capsule has defects. Specifically, the buoyancy signal features (such as the time-domain fluctuation amplitude, frequency-domain characteristic frequency) collected by the pressure sensor array and the image features (edge curvature, fluorescence intensity value) obtained by the optical imaging unit are input into the classification model, and compared and analyzed in combination with the preset thresholds. If it is determined to be a defective capsule, the control system sends an instruction to the rejection mechanism to reject it. This detection method comprehensively combines multi-dimensional detection data, and compared with the traditional single detection method, it realizes more accurate and efficient capsule defect detection.

[0042] The intelligent responsive pharmaceutical-grade silicone oil composition of the present invention realizes high-sensitivity detection of minute liquid leakage and fine cracks through the synergistic action of multiple components. For example, the sodium fluorescein derivative can specifically recognize hydroxyproline, the degradation product of gelatin, and can detect liquid leakage points at the level of 0.02 μL, with a lower detection limit than the traditional detection method. The temperature-sensitive polyethylene glycol-grafted silicone oil undergoes a sudden density change due to the temperature change of the liquid leakage, and combined with a high-precision pressure sensor, it can capture the liquid leakage signal more accurately. Compared with the traditional detection method, the present invention improves the liquid leakage detection sensitivity by several times, and the crack detection resolution reaches 0.1 mm, greatly improving the accuracy of defect detection.

[0043] Greatly enhanced compatibility: Traditional pharmaceutical-grade dimethicone is only applicable to the detection of gelatin capsules. For HPMC plant capsules with different densities, etc., separate liquids need to be configured, which is cumbersome to operate and costly. The intelligent responsive pharmaceutical-grade silicone oil composition of the present invention can continuously adjust the density in the range of 0.98 - 1.02 g / mL by adjusting the concentration of silica nanospheres, and can be compatible with more than 95% of commercially available capsules, including different types such as gelatin, HPMC, and pullulan polysaccharide. There is no need to frequently replace the detection liquid, which improves the versatility and detection efficiency of the detection equipment.

[0044] Effectively reduced detection cost: In terms of equipment cost, the detection equipment of the present invention is about 1 / 5 of the X-ray detection system. In terms of maintenance cost, due to the antibacterial effect of the triclosan-loaded nanospheres in the intelligent responsive pharmaceutical-grade silicone oil composition, the liquid circulation life cycle is extended from the traditional 500 times to 1000 times, and the maintenance cost is reduced by 70%. Moreover, the silicone oil replacement cost is <0.1 yuan / liter, and the raw material cost of the nanospheres only increases by 15%, but the comprehensive detection cost decreases by 30%. While ensuring the detection accuracy, the cost is effectively controlled.

[0045] Significantly improved detection reliability: The traditional detection method is easily interfered by external factors, resulting in misjudgment. The present invention adopts a multimodal detection technology, which integrates various information such as buoyancy signals, optical images, and fluorescence images for comprehensive judgment. For example, when detecting liquid leakage, not only relying on the buoyancy change, but also combining the fluorescence enhancement signal and the liquid seepage trace in the optical image. Through the machine learning algorithm built into the control system, the false alarm rate is greatly reduced, from the traditional 0.05% to 0.01%, improving the reliability of the detection results.

[0046] Example 1: Detection of liquid leakage in gelatin capsules.

[0047] Equipment parameter settings: The viscosity of the silicone oil is 50 cSt, and the density is 0.96 g / mL. The formula of the silicone oil composition is: 88% pharmaceutical-grade dimethicone, 5% silica nanospheres, 1.5% thermosensitive polyethylene glycol-grafted silicone oil, 0.3% fluorescein sodium derivative, and the balance is purified water impurities (meeting the density requirement of 0.96 g / mL). The circulation filtration frequency is 1 time / minute; the conveyor belt speed is 1 m / s (corresponding to a production capacity of 1200 capsules / minute); the pressure sensor accuracy is ±0.01 kPa, and the sampling frequency is 200 Hz; the optical imaging parameters are: the red light wavelength is 650 nm, and the camera frame rate is 1000 fps.

[0048] Detection process: The gelatin capsules to be inspected are arranged directionally by a vibrating disk and then fall onto a mesh belt conveyor. The capsules are immersed in silicone oil along with the conveyor belt. Since the density of the gelatin capsule (1.01 g / mL) is slightly greater than that of the silicone oil (0.96 g / mL), normal capsules are in a semi-floating state with an immersion depth of about 2 / 3. Leaking capsules will gradually sink due to the leakage of liquid medicine, resulting in an increase in the overall density. During the detection process, the pressure sensor array captures the buoyancy changes of the capsules in real time. The buoyancy fluctuation of normal capsules is small, generally ≤ ±0.05 kPa, while the buoyancy fluctuation of leaking capsules will increase significantly, exceeding 0.2 kPa. At the same time, a high-speed camera synchronously collects the surface images of the capsules. The high-speed camera synchronously collects visible light images and 700 nm fluorescence images. Leaking capsules show spots with an intensity > 3 times the reference value in the fluorescence channel due to the reaction of sodium fluorescein derivatives with hydroxyproline, forming a double verification with the buoyancy signal. The control system extracts and analyzes the features of the collected buoyancy signals and optical images, calculates the standard deviation of the buoyancy signals. If the standard deviation > 0.1 kPa and the optical image shows traces of liquid leakage, it is determined as a leaking capsule, and the rejection instruction is triggered. Qualified capsules smoothly enter the discharge port, and defective capsules are rejected by pneumatic nozzles.

[0049] Verification data: 10,000 gelatin capsules with 0.1 μL leakage holes were inspected, and 9,998 were correctly identified, with 2 misjudgments (both false positives caused by slight differences in filling amount), and the accuracy rate was as high as 99.98%. After continuous operation for 8 hours, the silicone oil pollution degree (measured by TOC) < 5 ppm, meeting the cleanliness requirements for drug production.

[0050] Example 2: Detection of cracks in plant capsules.

[0051] Equipment parameter adjustment: Adjust the density of silicone oil to 1.03 g / mL (by mixing 10% high-density silicone oil) to match the density of HPMC plant capsules (1.03 g / mL); the mesh belt conveyor is immersed 5 - 10 mm below the silicone oil surface to carry the capsules and move at a speed of 1 m / s; the accuracy of the pressure sensor is ±0.01 kPa, and the sampling frequency is 200 Hz; optical imaging parameters: red light wavelength 650 nm, camera frame rate 1000 fps.

[0052] Detection process: The HPMC plant capsules to be inspected are arranged directionally by a vibrating disk and then sent onto the conveyor belt. Normal HPMC plant capsules are in a suspended state in silicone oil with a density of 1.03 g / mL. However, cracked capsules will gradually sink due to the leakage of internal air and the resulting decrease in buoyancy. The red light laser of the optical imaging unit projects parallel light onto the capsule surface, and the high-speed camera collects 360° images. The control system analyzes the images by combining the background subtraction algorithm to identify cracks of 0.1 mm level on the capsule surface. If cracks are detected on the capsule surface and the buoyancy shows a downward change, it is determined as a defective capsule, and the rejection mechanism is activated to reject it.

[0053] Verification data: 1000 HPMC plant capsules with 0.1 mm - level cracks were tested, and the detection accuracy reached 99.7%.

[0054] Example 3: Detection of cracks in HPMC capsules by the intelligent - responsive pharmaceutical - grade silicone oil composition

[0055] Liquid formulation and equipment parameters: The liquid formulation is 91% pharmaceutical - grade dimethyl silicone oil (50 cSt), 6% silica nanospheres (density adjusted to 1.02 g / mL), 2% thermosensitive polyethylene glycol - grafted silicone oil, 0.2% sodium fluorescein derivative, and the remaining 0.8% are trace impurities introduced during the production process. The conveyor belt speed is 1.5 m / s, the sampling frequency of the pressure sensor is 500 Hz, and the fluorescence channel gain of the camera is 20 dB.

[0056] Detection process: Normal HPMC capsules (density 1.03 g / mL) are in a semi - floating state in this intelligent - responsive pharmaceutical - grade silicone oil composition, with an immersion depth of 70%. When there are cracks in the capsules, the internal gas leaks, increasing the density, and the sinking distance within 1 second is > 10 mm. At the same time, the sodium fluorescein derivative in contact with the leak hole reacts specifically with the hydroxyproline in the capsule to form a bright fluorescent spot with a diameter of 0.1 mm (intensity reference value 150%). The thermosensitive component will also cause local temperature changes due to liquid leakage, triggering a buoyancy pulse (amplitude 0.3 mN, duration 0.2 s). The control system comprehensively analyzes the buoyancy signal, optical image, and fluorescence signal collected through a multi - modal model (SVM + fluorescence intensity threshold) to determine whether there are crack defects in the capsules. If it is determined to be defective, the rejection mechanism will remove it.

[0057] Verification data: 1000 HPMC capsules with 0.02 μL leak holes were tested. 999 capsules were correctly identified, and only 1 capsule was misjudged due to the short - term deposition of nanospheres, which is a 28.5% improvement compared to the traditional silicone oil solution (700 capsules correctly identified).

[0058] Example 4: Detection of poor sealing of gelatin capsules

[0059] Equipment parameter settings: The viscosity of the silicone oil is 50 cSt, the density is 0.98 g / mL, the circulation filtration frequency is 1 time / minute; the conveyor belt speed is 1.2 m / s; the accuracy of the pressure sensor is ±0.01 kPa, and the sampling frequency is 300 Hz; optical imaging parameters: the red light wavelength is 650 nm, and the camera frame rate is 1200 fps.

[0060] Detection process: After the gelatin capsules are oriented by the vibrating disk, they enter the silicone oil detection liquid on the conveyor belt. For gelatin capsules with poor sealing, a small amount of air or contents will slowly leak. The pressure sensor array can capture the subtle buoyancy changes caused by the leakage, and its fluctuation characteristics are different from those of normal capsules. For example, the fluctuation frequency will increase, and the amplitude shows irregular changes within a certain period. The optical imaging unit observes whether there are traces of content leakage or abnormal gaps at the capsule joints by collecting the surface images of the capsules. The control system uses the SVM algorithm to analyze the buoyancy signal and the optical image data. When the change in the fluctuation frequency of the buoyancy signal exceeds a certain threshold (such as the frequency increases by more than 5 Hz), and the optical image shows suspected content leakage or the gap width exceeds 0.05 mm, it is determined as a capsule with poor sealing, and the rejection mechanism will remove it.

[0061] Verification data: 8000 gelatin capsules with potential sealing problems of different degrees were detected, 7980 were correctly identified, and 20 were misjudged, with an accuracy rate of 99.75%.

[0062] Example 5: Detection of leakage of pullulan capsules.

[0063] Equipment parameter adjustment: According to the density of pullulan capsules (about 1.00 g / mL), the density of silicone oil was adjusted to 1.00 g / mL by controlling the concentration of silica nanospheres; the conveyor belt speed was 0.8 m / s; the pressure sensor accuracy was ±0.01 kPa, and the sampling frequency was 400 Hz; the optical imaging parameters were: red light wavelength 650 nm, and the camera frame rate was 1500 fps.

[0064] Detection process: After the pullulan capsules are oriented by the vibrating disk, they enter the silicone oil along with the conveyor belt. Normal capsules are in a suspended or slightly floating state in the silicone oil with a matching density. Once leakage occurs, due to the density difference of the liquid medicine, the overall density of the capsule changes, and the pressure sensor array detects abnormal buoyancy fluctuation signals, such as the buoyancy fluctuation amplitude exceeding ±0.1 kPa within a short time. At the same time, the sodium fluorescein derivative will react with the leaked components (if the leaked liquid contains reactive substances), and the fluorescence channel of the high-speed camera can capture the change in fluorescence intensity. The images collected by the optical imaging unit can also assist in judging whether there are leakage traces. The control system synthesizes these data. When the buoyancy fluctuates abnormally and the change in fluorescence intensity exceeds 30% of the reference value, or the optical image shows obvious leakage signs, it is determined as a leaking capsule, which is removed by the rejection mechanism.

[0065] Verification data: 5000 pullulan capsules with 0.08 μL leakage holes were detected, 4990 were accurately identified, and 10 were misjudged, with an accuracy rate of 99.8%.

[0066] Example 6: Batch detection of mixed-type capsules.

[0067] Device parameter settings: Configure the density of silicone oil to be 1.00 g / mL (intermediate adaptation density, accommodating multiple types of capsules); conveyor belt speed 1 m / s; pressure sensor accuracy ±0.01 kPa, sampling frequency 500 Hz; optical imaging parameters: red light wavelength 650 nm, camera frame rate 2000 fps.

[0068] Detection process: After mixing different types of capsules such as gelatin, HPMC, and pullulan polysaccharide, they are preliminarily sorted by a vibratory bowl feeder and enter the conveyor belt. During the detection process, the pressure sensor array continuously collects buoyancy signals, and the optical imaging unit synchronously collects images. The control system classifies and analyzes the collected data based on the normal floating states and characteristic data of different types of capsules at this silicone oil density (obtained through previous training). For example, the change pattern of the buoyancy signal when a gelatin capsule leaks is different from that when an HPMC capsule leaks, and the control system uses the SVM algorithm to make a judgment based on these differences. When a defective capsule is judged, the rejection mechanism performs rejection according to the instructions of the control system.

[0069] Verification data: 10,000 mixed capsules (each type has a certain proportion and contains different defects) are detected. 9,950 defective capsules are correctly identified, and 50 are misjudged. The comprehensive accuracy rate reaches 99.5%. This fully verifies the effectiveness and accuracy of the system in batch detecting mixed-type capsules. Compared with the traditional method of separately detecting different types of capsules, it saves detection time and cost, and the detection efficiency is increased by approximately 40%.

[0070] Example 7: Verification of the influence of the full range of conveyor belt speeds (0.2 - 2 m / s) and immersion depths (5 - 10 mm) on the detection performance.

[0071] Test conditions

[0072] Detection object: Gelatin capsules (density 1.0 g / mL, 5000 non-defective samples, 100 artificially manufactured leaking defect samples, 100 cracked defect samples).

[0073] Silicone oil composition: 88% pharmaceutical-grade dimethyl silicone oil, 5% silicon dioxide nanospheres, 1.5% thermosensitive polyethylene glycol-grafted silicone oil, 0.3% fluorescein sodium derivative, and the balance is impurities, density 1.0 g / mL.

[0074] Detection container: The inner wall is coated with a polytetrafluoroethylene coating, and a microfluidic diversion groove is set.

[0075] Test process and results

[0076] The immersion depth of the conveyor belt is fixed at 5 - 10 mm, and the detection performance at different speeds is tested:

[0077] Speed 0.2 m / s: The conveyor belt is immersed 5 mm below the silicone oil liquid level (taking the boundary value), the sampling frequency of the pressure sensor is 100 Hz, and the frame rate of the optical imaging unit is 500 fps. The system processes 1000 capsules per hour (calculation: 0.2 m / s × 3600 s = 720 m / h, the width of the mesh belt is 0.1 m, the capsule spacing is 0.02 m, throughput = 720 m / h × 0.1 m / 0.02 m = 3600 capsules / h, actually adjusted to 1000 capsules / h due to start-stop). Defect recognition results: The recognition rate of liquid leakage defects is 99.8% (99 / 100), the recognition rate of crack defects is 99.6% (99 / 100), and the false alarm rate is 0.1% (5 defect-free samples misjudged).

[0078] Speed 1.0 m / s (reference speed): The conveyor belt is immersed 7 mm below the silicone oil liquid level (intermediate value), the sampling frequency of the pressure sensor is 100 Hz, and the frame rate of the optical imaging unit is 500 fps (default parameters). The system processes 5000 capsules per hour (the same as in Example 1). Defect recognition results: The recognition rate of liquid leakage defects is 99.9% (100 / 100), the recognition rate of crack defects is 99.8% (99 / 100), and the false alarm rate is 0.05% (3 defect-free samples misjudged).

[0079] Speed 2.0 m / s (high-speed boundary): The conveyor belt is immersed 10 mm below the silicone oil liquid level (taking the boundary value). To match the high-speed acquisition requirements, the sampling frequency of the pressure sensor is increased to 150 Hz, and the frame rate of the optical imaging unit is increased to 800 fps. The system processes 10000 capsules per hour (calculation: 2.0 m / s × 3600 s = 7200 m / h, the width of the mesh belt is 0.1 m, the capsule spacing is 0.02 m, throughput = 7200 m / h × 0.1 m / 0.02 m = 36000 capsules / h, actually adjusted to 10000 capsules / h due to detection time limitations). Defect recognition results: The recognition rate of liquid leakage defects is 99.2% (99 / 100), the recognition rate of crack defects is 99.0% (99 / 100), and the false alarm rate is 0.2% (10 defect-free samples misjudged).

[0080] Conclusion

[0081] Rationality of the conveyor belt speed range (0.2 - 2 m / s):

[0082] At low speed (0.2 m / s), although the processing efficiency is low, the sensor can fully collect data, and the defect recognition accuracy is extremely high; at high speed (2.0 m / s), by increasing the sensor sampling frequency and imaging frame rate, a defect recognition rate of more than 99% can still be maintained, verifying the technical feasibility of the speed range.

[0083] Necessity of the conveyor belt immersion depth (5 - 10 mm):

[0084] Regardless of the speed, when the conveyor belt is immersed 5 - 10 mm below the silicone oil surface, the capsules are completely immersed in the silicone oil, and the buoyancy is stable (the signal fluctuation of the pressure sensor ≤ ±0.01 kPa), avoiding the buoyancy detection error caused by partial exposure of the capsules above the liquid surface and ensuring the collaborative detection effect of the pressure sensor and the optical imaging unit.

[0085] Example 8: Verification of the detection performance of the sampling frequency of 100 Hz, the camera frame rate of 2000 fps, and the power parameters.

[0086] Test conditions

[0087] Detection object: Pullulan polysaccharide capsules (density 1.01 g / mL, 6000 defect-free samples, 150 samples with artificially manufactured poor sealing defects, and 150 samples with micro-crack defects).

[0088] Detection equipment parameters:

[0089] Pressure sensor array: The sampling frequency is set to 100 Hz, the detection accuracy is ±0.01 kPa, and it can capture the buoyancy changes caused by liquid leakage at the 0.05 μL level. Optical imaging unit: The high-speed camera frame rate is 2000 fps, the resolution is 1920×1080, and it is equipped with a red light laser (power 80 mW, wavelength 650 nm) to ensure clear imaging of fast-moving capsules and capture defects at the 0.1 mm level. Conveyor belt: The speed is set to 1.2 m / s, immersed 8 mm below the silicone oil surface, and the silicone oil composition is the same as the formula in Example 1 (density 1.01 g / mL to match the capsules).

[0090] Test process

[0091] Start the detection system. The pressure sensor collects the buoyancy signals of the passing capsules in real time at a frequency of 100 Hz, and the optical imaging unit takes pictures of the capsules at a frame rate of 2000 fps. The laser continues to work (power 80 mW) to provide a stable light source. The capsules pass through the detection container with the conveyor belt, and the control system synchronously processes the pressure signals (analyzing time-domain fluctuations and frequency-domain characteristics) and image data (edge curvature, gray-scale distribution, fluorescence intensity). Every 1000 capsules are detected, and the defect recognition results and the system operation status (such as the heating condition of the laser and the stability of the sensor) are recorded.

[0092] Test results

[0093] Defect recognition rate: The recognition rate of poor sealing defects is 99.7% (149 / 150), the recognition rate of micro-crack defects is 99.3% (149 / 150), and the false alarm rate of defect-free samples is 0.08% (5 misjudgments out of 6000). Equipment stability: The laser operates continuously for 8 hours at a power of 80 mW, with normal heat generation (housing temperature ≤ 45°C) and no light intensity attenuation; at a sampling frequency of 100 Hz for the pressure sensor, the signal transmission is stable and there is no data loss; the high-speed camera operates at a frame rate of 2000 fps, with no image smear and normal storage.

[0094] Conclusion

[0095] When the sampling frequency of the pressure sensor is 100 Hz, it can accurately capture the changes in buoyancy signals caused by capsule liquid leakage, etc., providing reliable data for defect judgment. The optical imaging unit with a frame rate of 2000 fps combined with an 80 mW laser ensures clear imaging of high-speed moving capsules and effectively identifies micro-defects.

[0096] The detection system and detection method of the present invention can be seamlessly integrated into the capsule production line, realizing fully automated operations from capsule feeding, detection to the removal of defective capsules, greatly improving production efficiency. For pharmaceutical companies and CMO factories, it can effectively reduce the defective rate of drugs, improve product quality, and reduce the recall risk caused by drug defects, with significant economic benefits. At the same time, its technology helps enterprises gain an advantageous position in the market competition, promotes the upgrading of quality inspection technology in the entire pharmaceutical industry, and has broad industrial application prospects.

[0097] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification only illustrate the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements fall within the scope of protection required by the present invention. The scope of protection required by the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-modal defect detection system for capsule drugs, characterized in that, Comprising: A detection container filled internally with an intelligent responsive pharmaceutical-grade silicone oil composition, the composition comprising by mass percentage: 88 - 92% of a pharmaceutical-grade dimethyl silicone oil with a viscosity of 50 cSt and a density of 0.96 g / mL; 3 - 6% of hydroxylated silica nanospheres with a particle size of 50 nm; 1 - 2% of a thermosensitive polyethylene glycol grafted silicone oil with a PEG molecular weight of 2000 Da; 0.1 - 0.5% of a sodium fluorescein derivative; The sum of the mass percentages of each component is 100%; A transmission system, comprising a horizontal mesh belt conveyor, the conveyor being immersed 5 - 10 mm below the silicone oil liquid level and moving at a constant speed of 0.2 - 2 m / s to carry capsules; A detection mechanism, including a pressure sensor array disposed at the bottom of the detection container and optical imaging units symmetrically disposed on both sides of the conveyor; A control system, electrically connected to the pressure sensor array and the optical imaging units, and built-in with a machine learning algorithm to fuse detection data to judge capsule defects; A rejection mechanism, disposed at the end of the conveyor, to reject defective capsules according to the instructions of the control system; The density of the intelligent responsive pharmaceutical-grade silicone oil composition is dynamically adjusted within the range of 0.98 - 1.02 g / mL by the concentration of silica nanospheres, and the surface tension is 20 - 25 mN / m.

2. The multi-modal defect detection system for capsule drugs according to claim 1, wherein: The surface tension of the pharmaceutical-grade dimethyl silicone oil is 20 - 25 mN / m, the moisture content is <0.1%, and it is circulated and filtered through a 0.22 μm sterilizing filter.

3. The multi-modal defect detection system for capsule drugs according to claim 1, wherein: The pressure sensor array comprises at least 10 distributed piezoresistive sensors, with an accuracy of ±0.01 kPa and a sampling frequency of ≥100 Hz.

4. The capsule drug multi-modal defect detection system according to claim 1, characterized in that: The wavelength of the red light laser of the optical imaging unit is 650 nm, the frame rate of the high-speed camera is ≥500 fps, and the resolution is ≥1280×1024.

5. The capsule drug multi-modal defect detection system according to claim 1, wherein: The machine learning algorithm of the control system is a support vector machine model, and the pre-training data includes at least 100,000 buoyancy-image feature pairs of normal / defective capsules.

6. The multi-modal defect detection system for capsule drugs according to claim 1, wherein: The inner wall of the detection container is coated with a polytetrafluoroethylene coating, the surface roughness Ra ≤ 0.2 μm, the conveyor is made of 316L stainless steel, and the mesh hole diameter is 1 / 2 - 2 / 3 of the capsule diameter.

7. The multi-modal defect detection system for capsule drugs according to claim 1, wherein: The inner wall of the detection container is provided with a microfluidic diversion groove to guide the intelligent responsive pharmaceutical-grade silicone oil composition to form a stable laminar flow.

8. The capsule drug multi-modal defect detection system according to claim 1, wherein: The high-speed camera of the optical imaging unit is integrated with a 700 nm band-pass filter to collect the specific fluorescence image of the sodium fluorescein derivative to assist in defect judgment.

9. A multi-dimensional detection method for capsule drugs, which is implemented using the capsule drug multi-modal defect detection system described in any one of claims 1-8, characterized in that, Including the following steps: S1. Adjust the density of the intelligent responsive pharmaceutical-grade silicone oil composition according to the capsule type; S2. The pressure sensor array collects the buoyancy fluctuation signal, and the optical imaging unit collects the optical image; S3. The control system fuses and analyzes the data, and judges whether the capsule has liquid leakage, cracks or surface defects according to the preset threshold; S4. If it is determined to be a defective capsule, the rejection mechanism will reject it.

Images