Process abnormality monitoring system and method by analyzing the cycle time and temperature pattern of a mold using an AioT-based sensor module

Abstract

The present invention relates to a process abnormality monitoring system and method based on pattern analysis of cycle time and temperature of a molding die, which uses an IoT-based sensor module to detect abnormal signals using five statistical and machine learning techniques to monitor the condition of the mold and the production process and to determine whether there is an abnormality in product quality, wherein the five methods each independently detect abnormal signals, and if it is determined that an abnormality has occurred in a corresponding shot by two or more of the five methods, the corresponding shot is labeled as an outlier and managed.

Description

Technology Field

[0001] The present invention relates to a system and method for detecting abnormal signals using five statistical and machine learning techniques to monitor the condition of a mold and a production process, and thereby determining whether there is an abnormality in product quality, wherein the five methods each independently detect abnormal signals, and if two or more of the five methods determine that a shot has an abnormality, the shot is labeled as an outlier and managed using an IoT-based sensor module, and the system and method for monitoring process abnormalities by analyzing the cycle time and temperature patterns of a molding mold. Background Technology

[0002] We examine conventional technologies regarding quality control of plastic injection molded products and mold management produced using the mold related to the present invention.

[0003] The business model of the manufacturing industry, which was previously organized and operated by small-scale enterprises, is recently evolving into large-scale global business chains in line with the establishment of global networks and the globalization of related markets.

[0004] The relationship between OEMs (ordering companies) and suppliers (production executors) is shifting from small-scale, local connections to global and large-scale ones. Consequently, there are limitations to managing manufacturing processes using traditional spreadsheet-based production management methods that rely on worker records and data entry.

[0005] To address these issues, we are developing a method to manage manufacturing processes using molds, which are key production media in the manufacturing industry, and IoT devices. This approach involves mounting IoT devices on molds and applying sensor modules (e.g., mold counters, shot counters, IoT sensing modules, edge sensing modules, etc.) to acquire production information.

[0006] There is a growing need for technology to manage manufacturing processes by measuring mold temperature and cycle time during product production.

[0007] However, in the case of sensing devices employing conventional technology, accurate temperature measurement is difficult because other materials (e.g., plastic for the sensor module case, air, etc.) exist between the measuring entity measuring the temperature within the sensor module and the mold, resulting in heat loss and the inability to measure the temperature transmitted from the mold. Furthermore, efficient processing of information is difficult because the measured data (temperature and cycle time) is collected and analyzed manually.

[0008] Considering the problems of the prior art described above, the present invention provides an analysis / monitoring system and method capable of automatically analyzing changes in mold temperature and cycle time using AI and monitoring the manufacturing process of an injection-molded product in real time. Prior art literature

[0009] Korean Registered Patent Publication No. 10-2274389 (B1) Korean Registered Patent Publication No. 10-2574092 (B1) The problem to be solved

[0010] The problem that the present invention aims to solve is to improve the inefficiency and information asymmetry of large-scale product production management. To solve these problems, the present invention provides a system and method for monitoring process abnormalities by analyzing the cycle time and temperature patterns of a mold for molding using an IoT-based sensor module that continuously detects cycle time and temperature changes during product production using IoT sensors and analyzes them in real time to increase production efficiency.

[0011] Another problem that the present invention aims to solve is the issue of failing to detect product quality variations early. In particular, since cycle time and temperature have a significant impact on product quality, it is necessary to analyze and manage them immediately. The present invention provides a system and method for monitoring process abnormalities based on pattern analysis of cycle time and temperature of a molding die, utilizing an IoT-based sensor module that analyzes and evaluates cycle time and temperature data collected during production in real time using statistical analysis and machine learning algorithms to detect and address quality issues that may occur during production at an early stage.

[0012] Another problem that the present invention aims to solve is that, in order to efficiently operate a mass production line, a management system is required that can remotely monitor and control the status of all processes without directly inspecting them on-site. The invention provides a process abnormality monitoring system and method based on pattern analysis of cycle time and temperature of a molding die using an IoT-based sensor module that detects abnormal signals in temperature and cycle time in real time using a sensor mounted on the sensor module and can remotely manage the situation when a problem occurs.

[0013] Another problem that the present invention aims to solve is to ensure consistency and accuracy in quality control. Since various variables occurring during the production process affect product quality, the invention provides a system and method for monitoring process abnormalities by analyzing patterns of cycle time and temperature of a molding die using an IoT-based sensor module. This is achieved by utilizing statistical algorithms installed on a server and AI-based abnormal signal detection techniques to precisely analyze the production status, combining multiple detection methods to increase the accuracy of quality control, thereby maintaining a normal production flow and enabling the consistent production of high-quality products.

[0014] The final problem that this invention aims to solve is the efficiency of mold management, which is a high-value asset. Since molds are expensive and stored and operated by external manufacturers, the invention provides a process abnormality monitoring system and method based on the analysis of cycle time and temperature patterns of molding molds using an IoT-based sensor module. This allows for the real-time identification of mold locations and real-time verification of abnormal conditions, thereby minimizing manpower and costs for mold management and unifying the management of molds installed worldwide. means of solving the problem

[0015] The means for solving the problem of the present invention is to provide a method for monitoring process abnormalities by pattern analysis of cycle time and temperature of a molding die using an IoT-based sensor module, comprising the steps of: measuring cycle time and temperature data from a sensor embedded in a sensor module installed in a mold; transmitting the measured cycle time and temperature data to a management server or server via a terminal using wireless or wired communication means; storing the transmitted cycle time and temperature data in the memory or database of the management server or server; pattern analysis of a product manufacturing process based on the cycle time and temperature data stored in the management server or server equipped with statistical analysis and machine learning algorithms; and early detection of quality problems regarding products that deviate from a normal pattern during the manufacturing process as a result of the pattern analysis of the cycle time and temperature data.

[0016] Another means of solving another problem of the present invention is to implement a process abnormality monitoring system based on pattern analysis of cycle time and temperature of a molding die using an IoT-based sensor module that includes a sensor embedded in a sensor installed on one side of a mold to measure cycle time and temperature data, transmits the measured cycle time and temperature data to a management server or server via a terminal using wireless or wired communication means, stores the transmitted cycle time and temperature data in the memory or database of the management server or server, and includes a management server or server equipped with statistical analysis and machine learning algorithms for pattern analysis of the product production process based on the cycle time and temperature data, and can detect quality problems early on for products that deviate from a normal pattern during the production process as a result of pattern analysis of the cycle time and temperature data.

[0017] Another means of solving another problem of the present invention is to provide a method for monitoring process abnormalities by analyzing the cycle time and temperature patterns of a molding die using an IoT-based sensor module, comprising the steps of calculating the mean and standard deviation of set cycle time data, subtracting the mean from each data point and dividing it by the standard deviation to calculate a Z-score, and monitoring values ​​that deviate from the normal data distribution through statistical analysis during the process of a manufactured product based on the Z-score.

[0018] Another means of solving the problem of the present invention is to implement a process abnormality monitoring system based on pattern analysis of cycle time and temperature of a molding die using an IoT-based sensor module that calculates the mean and standard deviation of set cycle time data, subtracts the mean from each data point and divides it by the standard deviation to calculate a Z-score, and monitors values ​​deviating from the normal data distribution through statistical analysis during the process of a product based on the Z-score.

[0019] Another means of solving another problem of the present invention is to provide a method for monitoring process abnormalities by analyzing the cycle time and temperature patterns of a molding die using an IoT-based sensor module, comprising the steps of calculating the median and quartiles of cycle time data, calculating the difference between the first quartile and the third quartile to calculate the quartile range value, and setting a lower limit and an upper limit by applying a specific weight to the quartile range value. Effects of the invention

[0020] The present invention contributes to minimizing information asymmetry between the original equipment manufacturer (OEM) and the producer. The OEM can check production process data in real time and track its history, so it does not have to rely solely on the information provided by the producer. In other words, the cycle time and mold temperature reported by the producer can be independently verified as actual data, and thus product quality issues can be proactively detected and responded to, thereby having an enhanced effect of increasing trust between the OEM and the producer.

[0021] Another advantage of the present invention is that it can reduce the time and cost required for analyzing and responding to quality issues by monitoring the production process in real time. Furthermore, when an abnormal signal is detected through an AI-based analysis algorithm, it can rapidly diagnose the cause of the problem and provide a solution, thereby minimizing production stoppages or the occurrence of defective products and resulting in cost reduction.

[0022] Another effect of the present invention is that cycle time and mold temperature are key factors in product quality during the injection process, and by collecting and analyzing these data in real time, the process stability is maintained to ensure product homogeneity. Additionally, by generating a warning in real time when the set temperature and cycle time range are exceeded or fell short, quality problems are prevented in advance, thereby reducing the defect rate and increasing the reliability of the final product.

[0023] Another advantage of the present invention is that it maximizes the efficiency of mold management and maintenance by automatically collecting mold data and detecting abnormal signals in real time through AI-based analysis, thereby enabling predictive maintenance. For example, by detecting changes in mold temperature or abnormalities in cycle time to predict and manage the mold condition, it has the advantageous effect of extending the lifespan of the mold and reducing unnecessary maintenance costs.

[0024] Another effect of the present invention is that it increases operational efficiency based on real-time data and AI solutions, and this technology can be utilized in various manufacturing processes beyond injection molding processes, thereby contributing to the synergistic effect of building a smart manufacturing foundation for the era of the Fourth Industrial Revolution across the entire manufacturing industry. Brief explanation of the drawing

[0025] Figure 1 is a flowchart illustrating the process of collecting and analyzing data in real time using the iMoldino IoT sensor. Figure 2 illustrates an algorithm procedure for detecting abnormal signals for each shot using cycle time and temperature data. Figure 3 illustrates the distinction between outliers representing abnormal signals and normal cycle times in clustered cycle times using the DBSCAN technique. Figure 4 illustrates the distinction between outliers and normal cycle times based on decision scores using the isolation forest algorithm. Figure 5 shows a sensor module with a built-in sensor for measuring cycle time and temperature data used in the present invention. Specific details for implementing the invention

[0026] We will examine the specific details for implementing the present invention.

[0027] The present invention relates to a process abnormality monitoring system and method based on pattern analysis of cycle time and temperature of a molding die, which uses an IoT-based sensor module to detect abnormal signals using five statistical and machine learning techniques to monitor the condition of the mold and the production process and to determine whether there is an abnormality in product quality, wherein the five methods each independently detect abnormal signals, and if it is determined that an abnormality has occurred in a corresponding shot by two or more of the five methods, the corresponding shot is labeled as an outlier and managed.

[0028] The sensor module (502 in FIG. 5) of the specification of the present invention is composed of a 'magnetic part (503 in FIG. 5)' and a 'sensing part (501 in FIG. 5)', and the sensing part is equipped with a Hall sensor, a temperature sensor, a vibration sensor, a measuring PCB, and a control PCB.

[0029] The sensor module (502 in FIG. 5) is shown in FIG. 5 for reference, and the magnetic part and the sensing part are separately fixedly installed on the movable side mold or the fixed side mold, respectively, so that the cycle time can be measured by measuring the mold opening or closing signal of the mold from the Hall sensor.

[0030] The hardware configuration of the sensor module is described in more detail in Korean Registered Patent Publications No. 10-2274389 (B1) and No. 10-2574092 (B1) registered by the applicant of the present invention.

[0031] More specifically, the sensor module used in the present invention is attached to the mold in use at the B-1 factory of manufacturer B of company A, and the sensor module continuously collects and transmits cycle time and temperature data generated during the injection process.

[0032] This data is measured and stored whenever each shot is completed, and the data measured from the corresponding mold is periodically transmitted to a terminal within the factory.

[0033] The terminal is connected to the outside world via a Wi-Fi or LTE communication module installed on-site, and data is transmitted between the mold's sensors and the terminal using Bluetooth communication.

[0034] The communication means employed in the present invention may selectively employ conventional wired or wireless communication means as needed.

[0035] The accumulated data is automatically transmitted to the S3 storage system according to the present invention at the site.

[0036] This process is carried out through an automated data transmission system, allowing users to securely transmit data to a cloud server without any separate manual work. Data stored in S3 is refined through the ETL (Extract, Transform, Load) process according to the present invention and converted into a format suitable for analysis. The ETL process includes preprocessing steps such as handling missing values, data normalization, and removing duplicate data to secure data optimized for subsequent analysis.

[0037] Subsequently, refined cycle time and temperature data for each mold are transmitted to a management server equipped with an abnormal signal detection algorithm. The management server monitors abnormal signals during the process using five statistical and machine learning techniques.

[0038] The methods used here include Z-score based, Interquartile Range (IQR) based, temperature analysis, DBSCAN, and Isolation Forest; each technique analyzes the data independently, and a shot is determined to be an outlier if multiple techniques detect an anomaly.

[0039] The analysis results from the management server or the server are generated in the form of a table, and the generated table is then recorded and stored in the management server or the server's database.

[0040] The database tables recorded and stored in the management server or the server's database contain information regarding shot abnormalities by mold, the time of occurrence, and the cause of the detected abnormal signal (e.g., cycle time or temperature). These recorded database tables are utilized for analysis and recording purposes based on database scan algorithms and play an important role in the subsequent monitoring process.

[0041] The web solution in the present invention provides a monitoring dashboard that provides the user with the status of the mold in real time. The dashboard provided in the present invention is designed to allow the user to grasp the product production status, cycle time, and temperature fluctuations of each mold at a glance.

[0042] In addition, the present invention is configured to allow the setting of a user-specified notification threshold, thereby providing a function that immediately notifies the user when the mold status deviates from a specific standard. Through this, the user can remotely monitor the mold status in real time and take appropriate preventive measures in advance if abnormal signals from the mold occur frequently.

[0043] Figure 1 shows the process of real-time data collection and analysis using IoT sensors attached to a mold, and the overall flow of data used to detect abnormal conditions.

[0044] In the first step, IoT sensors attached to the mold are used to collect cycle time and temperature data in real time during the injection molding process. This data provides a foundation for identifying variations in the injection process and detecting anomalies at an early stage.

[0045] In the second step, the collected data is transmitted from the sensor attached to the mold to the field terminal. The terminal serves as a temporary data storage facility and can perform simple preprocessing if necessary.

[0046] After simple preprocessing, transfer the data to the management server, server memory, or database.

[0047] In the third step, data transmitted from each terminal is sent to a management server, the server's memory, or a database, where it is managed in a single integrated form. The aforementioned management server, the server's memory, or the database collectively stores injection molding process data for all molds and manages it so that it can be utilized for various analyses.

[0048] In the fourth step, data stored in the management server, server memory, or database is analyzed using statistical techniques and machine learning-based anomaly detection algorithms. These algorithms detect specific data that deviates from the normal injection molding process pattern, and if an abnormal state is detected, they visually display it or issue a warning message.

[0049] Finally, the entire production process is monitored based on the results of the anomaly detection algorithm. If abnormal data is detected, field engineers can verify the data and take immediate preventive measures. This real-time monitoring system provides a synergistic effect that improves the quality of the production process and minimizes the defect rate.

[0050] Figure 2 shows the step-by-step process of detecting abnormal signals in the injection molding process using statistical techniques and machine learning algorithms. First, the left side of the figure shows two types of data collected from sensors and stored in a management server or a server database.

[0051] One is data on cycle time, and the other is temperature data. These data are preprocessed into a format capable of detecting abnormal data patterns by applying statistical techniques and machine learning algorithms.

[0052] Cycle time, temperature, and time data monitored during the injection process according to the present invention detect abnormal signals using four algorithms.

[0053] First, the z-score standardizes data and detects outliers by determining how far each data point is from the mean.

[0054] Second, the Interquartile Range (IQR) method refers to the interquartile range of the data distribution and identifies data points outside the range calculated by multiplying the IQR by a specific weight as outliers.

[0055] As shown in Figure 3, DBSCAN is a density-based clustering algorithm that distinguishes between dense and sparse regions of data and detects data in low-density areas as outliers. The parts marked with black dots in Figure 3 represent normal data, and these data represent normal values ​​formed into clusters by the DBSCAN algorithm.

[0056] The boundaries of this cluster were visually distinguished by adjusting the brightness, allowing the distribution of normal data to be clearly identified. On the other hand, the points marked in blue are data points outside the cluster and were classified as outliers by the DBSCAN algorithm. These outliers are data generated by abnormal situations, such as machine malfunctions, operator errors, or measurement sensor errors.

[0057] In addition, the data marked with purple dots represents the idle state, which corresponds to the time and interval during which the machine is not operating. As a result of the analysis in Figure 3, the DBSCAN algorithm clearly distinguished normal clusters based on cycle time data and detected data points located outside the clusters as outliers. This enables the rapid detection of abnormal machine conditions in manufacturing sites and contributes to improving operational efficiency.

[0058] As shown in Figure 4, the Isolation Forest detects outliers based on the degree of splitting within the tree structure. In this figure, black dots represent normal data (Inliers), and blue dots represent outliers. Additionally, purple dots represent the idle state.

[0059] The standby state refers to the period when the machine has stopped operating. The green line is a visual representation of the Decision Score, which is calculated as 1 - anomaly score. The higher the Decision Score value, the more the corresponding data point is determined to be normal data. In Figure 4, data located within the green boundary is classified as normal data, while data outside the boundary is determined to be an outlier.

[0060] Temperature data uses the z-score algorithm to evaluate how far each temperature measurement deviates from the normal range, thereby detecting abnormal signals.

[0061] The results derived from each algorithm are comprehensively evaluated by a majority algorithm. This prevents reliance solely on the results of individual algorithms and allows for the configuration of a final decision based on common anomaly patterns across multiple algorithms.

[0062] Finally, abnormal signals detected for each shot can be visually displayed or indicated as warning signals so that field engineers can recognize the situation.

[0063] The present invention proposes a method for detecting abnormal signals using five statistical and machine learning techniques to monitor the condition of a mold and the production process, and thereby determining whether there are quality defects in the product.

[0064] The above five methods each independently detect abnormal signals, and if two or three or more of the five methods determine that a shot has an abnormality, the shot is labeled as an outlier to manage the process.

[0065] The first method is a Z-score-based outlier detection method that utilizes statistical analysis to effectively detect values ​​deviating from a normal data distribution.

[0066] The above method calculates the mean and standard deviation of the given cycle time data, then calculates the Z-score by subtracting the mean from each data point and dividing the result by the standard deviation.

[0067] The Z-score calculated by the above method represents the standardized value of how far the data is from the mean, and generally, if the Z-score exceeds 3 or is less than -3, the data is considered an outlier that deviates from the normal value.

[0068] Z-score-based methods are simple yet efficient. Based on the mean and standard deviation of the data, they have low computational complexity and can be rapidly applied even to large datasets.

[0069] Furthermore, it is designed under the assumption that the data follows a normal distribution, providing statistical stability and reliability. This method is suitable for detecting outliers in cycle time data caused by mechanical defects, variations in operator proficiency, errors in measurement sensors, and the like.

[0070] Furthermore, the proportion of outliers identified through the Z-score method is used to set the contamination parameter of the Isolation Forest algorithm.

[0071] This contributes to optimizing the detection performance of the Isolation Forest algorithm by accurately reflecting the expected proportion of outliers within the dataset. The organic combination of these algorithms plays a crucial role in further enhancing the precision and reliability of the overall anomaly detection algorithm.

[0072] The second method is to detect outliers by utilizing the Interquartile Range (IQR). Quartile-based methods are designed to effectively detect values ​​that deviate from the median distribution of the data.

[0073] The quartile-based method includes a step of calculating the median and quartiles of the cycle time data, and a step of calculating the quartile range (IQR) value by finding the difference between the first quartile (Q1) and the third quartile (Q3).

[0074] Subsequently, it includes a step of setting a lower bound and an upper bound by applying specific weights to the interquartile range (IQR) values.

[0075] Here, the step of appropriately setting the weights used to extend the range of the interquartile range (IQR) according to the characteristics of the data may be included.

[0076] Generally, it is desirable to set the weight value to 1.5 or 3.

[0077] If the cycle time data is smaller than the lower limit or larger than the upper limit, the shot is considered an outlier outside the Interquartile Range (IQR).

[0078] Quartile-based methods are useful for detecting anomalies in data distribution that may occur due to mechanical defects, changes in worker behavior, or errors in measurement sensors.

[0079] In addition, quartile-based methods are not sensitive to data outliers and are robust against data skewness.

[0080] Therefore, when analyzing cycle time data collected in real time through the IoT sensor mounted on the sensor module of the present invention, quantitative and reliable outlier detection can be performed by applying the IQR method.

[0081] The third method is a temperature-based method. Since the sensor module of the present invention can collect temperature data simultaneously with the cycle time, it can determine whether there is an abnormality based on the temperature when producing the product by the corresponding shot.

[0082] Temperature is a very important indicator in the product manufacturing process, and temperatures outside the normal production temperature range can affect product quality.

[0083] Similar to the first method, temperature anomaly signals detect outliers by normalizing the temperature data and utilizing Z-scores.

[0084] The fourth method is to detect outliers by utilizing density-based clustering (DBSCAN, Density-Based Spatial Clustering of Applications with Noise).

[0085] Density-based clustering forms clusters based on the density between data and analyzes the number of data points (min_samples) included within a specific distance radius (eps) to distinguish high-density areas as clusters.

[0086] Through this method, data points deviating from the data cluster—that is, outliers—can be effectively detected. Cycle time data typically forms clusters within a certain radius under normal operating conditions. However, outliers that fall outside the normal range of clusters may occur due to factors such as machine malfunctions, differences in operator proficiency, or errors in measurement sensors.

[0087] Therefore, by utilizing cycle time data collected in real time from IoT sensors according to the present invention and applying a density-based clustering algorithm, such outliers can be automatically detected.

[0088] The density-based clustering algorithm was applied exclusively to the cycle time variable; this measure is intended to prevent the problem of misclassifying normal data as outliers in sections where shot times are missing.

[0089] In addition, the key parameters of the density-based clustering algorithm are adjusted to be dynamically changed according to the characteristics of real-time data.

[0090] Specifically, the radius distance (eps) is dynamically set by calculating the maximum distance (Euclidean) between data using the KNN (K-Nearest Neighbors) algorithm (k=2), and the minimum number of samples (min_samples) is configured to be automatically adjusted based on the number of data using a logarithmic function.

[0091] Figure 4 illustrates the result of applying a density-based clustering algorithm to cycle time data, and the visualization result is shown in one dimension (cycle time).

[0092] Figure 4 shows the visualization result of applying a density-based clustering algorithm to cycle time data collected from IoT sensors embedded in a sensor module.

[0093] In Figure 4, the parts marked with black dots represent normal data, and these data represent normal values ​​formed into clusters by a density-based clustering algorithm.

[0094] The boundary of this cluster was made visually distinguishable by adjusting the brightness, which allows the distribution of normal data to be clearly identified.

[0095] On the other hand, the points marked in blue are data points outside the cluster and are classified as outliers by density-based clustering algorithms. These outliers are data generated by abnormal situations, such as machine malfunctions, operator errors, or measurement sensor errors.

[0096] In addition, data marked with purple dots represents the idle state, which corresponds to the time and interval during which the machine is not operating. As a result of the drawing analysis, the density-based clustering algorithm clearly distinguishes normal clusters based on cycle time data and detects data points located outside the clusters as outliers.

[0097] This can contribute to rapidly detecting abnormal machine conditions at the manufacturing site and increasing operational efficiency.

[0098] The fifth method is to detect outliers using the Isolation Forest technique. While existing tree-based algorithms branch trees by calculating information gain based on entropy or Gini impurity, Isolation Forest identifies outliers based on how quickly each data point becomes isolated.

[0099] The core principle of the Isolation Forest algorithm is that while more branching processes are required to isolate data points when similar data is densely clustered nearby, outliers have low density and can be easily isolated with relatively few branches.

[0100] The Isolation Forest algorithm repeatedly branches the tree using arbitrary variables and arbitrary splitting points. Subsequently, it calculates the average number of branches required for a specific data point to be isolated within each tree and derives a final anomaly score based on this.

[0101] Through this, areas with high data density can be classified as normal data, while areas with low density can be classified as outliers.

[0102] In the case of cycle time data, data points form a dense distribution under normal operating conditions; however, outliers caused by machine malfunctions or operator errors have a low density, making them highly likely to be isolated into fewer branches by the Isolation Forest algorithm.

[0103] In this invention, utilizing these characteristics, the isolation forest algorithm is applied specifically to the cycle time variable. This is intended to prevent misclassification of normal data that may occur in sections where shot times are missing.

[0104] In addition, the system is designed to dynamically adjust key parameters of the isolation forest to respond to changes in the characteristics of data collected in real time. Specifically, the contamination parameter is configured to automatically adjust using the ratio of detected outliers based on the Z_Score. The attached figure shows the visualization results of applying the isolation forest algorithm to cycle time data and visually represents outlier detection in one dimension (cycle time).

[0105] Figure 4 shows the visualization result of applying the isolation forest algorithm to cycle time data collected from IoT sensors embedded in the sensor module.

[0106] In Figure 4, black dots represent normal data (Inlier), and blue dots represent outliers. Additionally, purple dots represent the idle state. The idle state refers to the period when the machine has stopped operating.

[0107] In Fig. 4, the green line visually represents the Decision Score, which is calculated as 1 minus the anomaly score. The higher the Decision Score value, the more the corresponding data point is determined to be normal data.

[0108] In Figure 4, data located within the green boundary is classified as normal data, and data outside the boundary is identified as outliers.

[0109] The five methods mentioned above are not applied uniformly to the entire cycle time data, but are applied only to specific time intervals (windows) by setting a specific window. The reason for setting such a window is that cycle time can be intentionally altered due to mold wear or the judgment of field engineers.

[0110] When fluctuating cycle times persist for an extended period, it is necessary to flexibly set abnormal signal detection criteria in consideration of this. Therefore, abnormal signal detection is performed by reflecting variability within the window, rather than using fixed criteria.

[0111] When an abnormal signal is determined in two or three or more of the five methods for monitoring the production process of a product according to the present invention and managing the quality, the corresponding shot is labeled as an abnormal value and managed, thereby enabling early detection of abnormalities that may occur during the production process and prompt response.

[0112] The present invention aims to enhance data-driven production quality management and cost efficiency in the injection molding process. The industrial applicability of this invention is extensive, and it is particularly applicable to most injection molding processes that utilize molds. By monitoring and analyzing temperature and cycle time data—the most critical factors in the injection molding process—in real time, quality issues can be prevented in advance, and process stability can be ensured.

[0113] a. Quality control and cost efficiency in the injection molding process

[0114] Variations in cycle time and mold temperature are factors that significantly affect the quality of the product. The present invention contributes to solving quality problems in the following cases.

[0115] Case 1: Increase in cycle time

[0116] Increased injection time, increased holding pressure time, increased cooling time, increased ejection time.

[0117] Problems: Increased production costs, increased possibility of ejection failure due to supercooling.

[0118] Quality issues: Possibility of poor surface condition, poor gloss, dimensional defects, etc.

[0119] Case 2: Cycle time reduction

[0120] Reduced injection time, reduced holding pressure time, reduced cooling time.

[0121] Problem: Production efficiency increases, but there is a possibility of reduced quality stability.

[0122] Quality issues: Warping, post-deformation, poor shrinkage, and dimensional defects.

[0123] Case 3: Mold temperature rise

[0124] Rise in cooling water temperature, cooling device malfunction, increase in mold internal scale.

[0125] Problems: Burrs, bending, warping, increased post-deformation, dimensional defects.

[0126] Quality issues: Possibility of poor shrinkage and product deformation.

[0127] Case 4: Mold temperature reduction

[0128] Coolant temperature drop, seasonal factors, cooling system malfunction.

[0129] Problems: Increased ejection defects, increased residual stress, reduced surface gloss.

[0130] Quality Issues: Increased warping and post-deformation.

[0131] b. Data utilization in the production process

[0132] Monitoring of temperature and cycle time compliance: OEMs and manufacturers can evaluate process quality stability through temperature and cycle time data and resolve quality issues in advance based on this.

[0133] Data-driven quality assessment: By utilizing data collected in real-time, quality issues can be identified immediately and the accuracy of reports verified without waiting for the producer's report.

[0134] Accurate production cost management: Quality and production unit costs can be managed simultaneously by utilizing cycle time and temperature data.

[0135] c. AI-based injection molding monitoring technology

[0136] The present invention refines quality control and problem cause analysis through AI-based monitoring technology, and enables manufacturing process innovation suitable for the era of the Fourth Industrial Revolution. This technology provides the following advantages.

[0137] Combination of Data and AI: Predicting and resolving quality issues in advance through the combination of real-time data and AI algorithms.

[0138] Increased trust between OEM and manufacturer: Establishing mutual trust by offsetting information asymmetry based on changes in cycle time and temperature.

[0139] Expanded applicability in manufacturing sites: Can be utilized as a foundational technology for applying AI monitoring technology to various manufacturing processes, not just injection molding.

[0140] d. Combined analysis of temperature and cycle time

[0141] The problems that can occur due to the combination of temperature and cycle time can be specifically summarized as shown in Table 1.

[0142]

[0144] Based on the above description, the scope of protection sought in the present invention is summarized, and the scope of protection of the present invention may be described in various modified ways as necessary.

[0145] The present invention comprises the steps of: measuring cycle time and temperature data from a sensor embedded in a sensor module installed in a mold; transmitting the measured cycle time and temperature data to a management server or a server via a terminal using wireless or wired communication means; storing the transmitted cycle time and temperature data in the memory or database of the management server or the server; analyzing the product production process pattern based on the cycle time and temperature data stored in the management server or the server equipped with statistical analysis and machine learning algorithms; and detecting quality problems early on for products that deviate from a normal pattern during the production process as a result of the pattern analysis of the cycle time and temperature data.

[0146] Another example of the present invention may include the step of calculating the mean and standard deviation of set cycle time data, subtracting the mean from each data point and dividing it by the standard deviation to calculate a Z-score, and the step of monitoring values ​​deviating from the normal data distribution through statistical analysis during the process of a manufactured product based on the Z-score.

[0147] Another example of the present invention may include the step of setting the data as an outlier outside the normal range if the calculated Z-score represents how far the data is from the mean as a standardized value, and if the Z-score exceeds 3 or is less than -3.

[0148] Another example of the present invention may include the steps of calculating the median and quartiles of cycle time data, calculating the difference between the first quartile and the third quartile to compute the quartile range value, and applying a specific weight to the quartile range value to set a lower limit and an upper limit.

[0149] Another example of the present invention may include the step of setting the weight value to 1.5 or 3.

[0150] Another example of the present invention may include a step of determining that the product is an outlier in the interquartile range when one or more of mechanical defects, changes in worker behavior, and errors in measurement sensors occur, and the cycle time data is smaller than the lower limit or larger than the upper limit.

[0151] Another example of the present invention may include the step of forming clusters based on the density between data in one or more of cycle time and temperature data, analyzing the number of data included within a specific distance radius to classify high-density areas into clusters, and using density-based clustering to determine data outside the clusters as outliers.

[0152] Another example of the present invention may include the step of monitoring outliers that fall outside the cluster corresponding to the normal range due to at least one of machine malfunction, differences in operator skill level, and errors in measurement sensors.

[0153] Another example of the present invention may include the step of providing a monitoring dashboard that services the status of the mold in real time to the user so that the product production status, cycle time, and temperature fluctuations in the mold can be identified at a glance.

[0154] Another example of the present invention provides a system comprising a sensor module with a built-in sensor installed on one side of a mold to measure cycle time and temperature data, a management server or server that transmits the measured cycle time and temperature data to a management server or server via a terminal using wireless or wired communication means, stores the transmitted cycle time and temperature data in the memory or database of the management server or server, and includes a management server or server equipped with a statistical analysis and machine learning algorithm for pattern analysis of a product manufacturing process based on the cycle time and temperature data, and is capable of early detection of quality problems regarding products that deviate from a normal pattern during the manufacturing process as a result of pattern analysis of the cycle time and temperature data.

[0155] Another example of the present invention provides a system that calculates the mean and standard deviation of set cycle time data, subtracts the mean from each data point and divides it by the standard deviation to calculate a Z-score, and monitors values ​​deviating from the normal data distribution through statistical analysis during the production process of a manufactured product based on the Z-score.

[0156] Another example of the present invention provides a system in which the calculated Z-score represents a standardized value indicating how far the data is from the mean, and if the Z-score exceeds 3 or is less than -3, the data is set as an outlier deviating from the normal value.

[0157] Another example of the present invention provides a system that calculates the median and quartiles of cycle time data, calculates the difference between the first quartile and the third quartile to compute a quartile range value, and sets a lower limit and an upper limit by applying a specific weight to the quartile range value. Industrial applicability

[0158] The AioT sensor and AI-based monitoring technology of the present invention is an optimal core technology capable of minimizing inefficiencies and asymmetry in manufacturing processes while simultaneously ensuring uniform quality and optimizing costs. Both producers and OEMs can respond immediately to quality issues through real-time data collected by sensors and processed by servers, and since it can be utilized as an essential foundational technology for smart manufacturing in the era of the Fourth Industrial Revolution, its industrial applicability is very high. Explanation of the symbols

[0159] 501; Sensing unit 502; Sensor module 503; Magnetic part

Claims

Claim 1 A method for monitoring process abnormalities by analyzing the cycle time and temperature patterns of a molding die using an IoT-based sensor module, comprising: a step of measuring cycle time and temperature data from a sensor embedded in a sensor module installed in the die; a step of transmitting the measured cycle time and temperature data to a management server or a server via a terminal using wireless or wired communication means; a step of storing the transmitted cycle time and temperature data in the memory or database of the management server or the server; a step of analyzing the product production process pattern based on the cycle time and temperature data stored in the management server or the server equipped with statistical analysis and machine learning algorithms; and a step of early detection of quality problems regarding products that deviate from a normal pattern during the production process as a result of the pattern analysis of the cycle time and temperature data. Claim 2 A method for monitoring process abnormalities by pattern analysis of cycle time and temperature of a molding die using an IoT-based sensor module, characterized in that, in claim 1, it includes the step of calculating the mean and standard deviation of set cycle time data, subtracting the mean from each data point and dividing it by the standard deviation to calculate a Z-score, and the step of monitoring values ​​deviating from a normal data distribution through statistical analysis during the process of a manufactured product based on the Z-score. Claim 3 A method for monitoring process abnormalities by analyzing the cycle time and temperature patterns of a molding die using an IoT-based sensor module, characterized in that, in claim 2, the calculated Z-score represents a standardized value indicating how far the data is from the mean, and if the Z-score exceeds 3 or is less than -3, the data is set as an outlier deviating from the normal value. Claim 4 A method for monitoring process abnormalities by analyzing the cycle time and temperature patterns of a molding die using an IoT-based sensor module, characterized in that, in claim 1, the method comprises the steps of calculating the median and quartiles of the cycle time data, calculating the difference between the first quartile and the third quartile to calculate the quartile range value, and setting a lower limit and an upper limit by applying a specific weight to the quartile range value. Claim 5 A method for monitoring process abnormalities by analyzing the cycle time and temperature patterns of a molding die using an IoT-based sensor module, characterized in that, in claim 4, the weight value is set to 1.5 or 3. Claim 6 A method for monitoring process abnormalities by analyzing the cycle time and temperature patterns of a molding die using an IoT-based sensor module, characterized in that, in claim 4, the product is determined to be an outlier in the quartile range when one or more of mechanical defects, changes in worker behavior, and errors in measurement sensors occur, and the cycle time data is smaller than the lower limit or larger than the upper limit. Claim 7 A method for monitoring process abnormalities by analyzing patterns of cycle time and temperature of a molding die using an IoT-based sensor module, characterized in that, in claim 1, it comprises the steps of forming clusters based on the density between data in one or more of cycle time and temperature data based on an isolation forest algorithm, analyzing the number of data included within a specific distance radius to classify high-density areas into clusters, and determining data outside the clusters as outliers using density-based clustering. Claim 8 A method for monitoring process abnormalities by analyzing the cycle time and temperature patterns of a molding die using an IoT-based sensor module, characterized in that, in claim 7, it includes the step of monitoring an abnormal value that deviates from a cluster corresponding to a normal range due to at least one of machine malfunction, differences in operator proficiency, and errors of a measurement sensor. Claim 9 A method for monitoring process abnormalities by analyzing the cycle time and temperature patterns of a molding die using an IoT-based sensor module, characterized in that, in claim 7, it includes the step of providing a monitoring dashboard that provides the user with the status of the mold in real time so that the product production status, cycle time, and temperature fluctuations in the mold can be grasped at a glance. Claim 10 A process abnormality monitoring system based on pattern analysis of cycle time and temperature of a molding die using an IoT-based sensor module, comprising: a sensor module having a built-in sensor installed on one side of the die to measure cycle time and temperature data; transmitting the measured cycle time and temperature data to a management server or server via a terminal using wireless or wired communication means; storing the transmitted cycle time and temperature data in the memory or database of the management server or server; and including a management server or server equipped with statistical analysis and machine learning algorithms for pattern analysis of the product manufacturing process based on the cycle time and temperature data, characterized by early detection of quality problems regarding products that deviate from a normal pattern during the production process as a result of pattern analysis of the cycle time and temperature data. Claim 11 A process abnormality monitoring system based on pattern analysis of cycle time and temperature of a molding die using an IoT-based sensor module, characterized in that, in claim 10, after calculating the mean and standard deviation of set cycle time data, the mean is subtracted from each data point and divided by the standard deviation to calculate a Z-score, and based on the Z-score, values ​​deviating from the normal data distribution are monitored through statistical analysis during the process of a manufactured product. Claim 12 A process abnormality monitoring system based on pattern analysis of cycle time and temperature of a molding die using an IoT-based sensor module, characterized in that, in claim 11, the calculated Z-score represents a standardized value indicating how far the data deviates from the mean, and if the Z-score exceeds 3 or is less than -3, the data is set as an outlier deviating from the normal value. Claim 13 A process abnormality monitoring system based on pattern analysis of cycle time and temperature of a molding die using an IoT-based sensor module, characterized in that, in claim 10, the median and quartiles of cycle time data are calculated, the difference between the first quartile and the third quartile is calculated to determine the quartile range value, and a lower limit and an upper limit are set by applying a specific weight to the quartile range value.

Images