Drawing-based automatic specification extraction and verification ordering system

Abstract

The present invention relates to a drawing-based automatic specification extraction and verification ordering system that automatically extracts dimensional, structural, and specification information from drawings and specification images of various facilities or devices including lighting fixtures, performs verification and correction by comparing the extracted information with a standard catalog, and automatically generates wiring connection guides and purchase orders.

Description

Technology Field

[0001] The present invention relates to a drawing-based automatic specification extraction and verification ordering system, and more specifically, to a drawing-based automatic specification extraction and verification ordering system that automatically extracts dimensional, structural, and specification information from drawings and specification images of various facilities or devices including lighting fixtures, performs verification and correction by comparing the extracted information with a standard catalog, and automatically generates wiring connection guidance and order forms. Background Technology

[0003] Lighting fixtures, particularly various lighting devices including luminaires, are installed in diverse environments such as buildings, industrial facilities, commercial facilities, and residential spaces. Depending on the installation environment and intended use, numerous specification elements must be comprehensively considered, including dimensions, height, hole pattern, luminous flux, color temperature, power consumption, power supply method, wiring structure, and the inclusion of mounting brackets and accessories. Typically, the design, ordering, and construction of such lighting fixtures are carried out based on drawings and specifications; the ordering manager or designer reviews these drawings and specifications to select suitable products and prepare purchase orders.

[0004] However, in actual work environments, drawings and specifications for lighting fixtures are often provided in the form of image files or scanned documents. Furthermore, since drawing notation methods and specification formats vary by manufacturer, there is a problem in that consistent interpretation requires significant time and expertise. In particular, unique graphic elements in drawings, such as dimension lines, arrows, reference points, and symbols, carry structural meanings unlike simple textual information; therefore, failure to interpret them accurately significantly increases the likelihood of dimensional errors or installation incompatibility.

[0005] Traditionally, the method of manually examining drawings or specifications to extract necessary information and draft purchase orders has been commonly used. Some automation technologies have proposed methods that utilize character recognition to extract text from specifications or input values ​​into predefined fields. However, since these conventional technologies rely primarily on character recognition or simple data input, they have limitations in accurately identifying structural information, such as dimensions, heights, and hole patterns, from drawing images. Consequently, problems arise where dimension values ​​are inaccurately recognized or reliability is compromised due to factors such as scan resolution, magnification or reduction ratios, and drawing distortion.

[0006] Furthermore, conventional technology often relies solely on a simple comparison of values ​​extracted from drawings or specifications with standard catalogs, failing to adequately reflect differences in tolerance ranges or notation methods among manufacturers. Consequently, errors may occur where products that are actually usable are judged as merely inconsistent, or conversely, products outside the acceptable range are deemed suitable. These issues lead to judgment errors during the ordering stage, causing reordering, schedule delays, and increased costs.

[0007] Furthermore, while information related to wiring connections is typically provided in the form of wiring diagrams or symbols within drawings, conventional technology fails to systematically interpret these symbols to automatically provide guidance on terminal locations, polarity, and grounding status. Consequently, on-site workers must interpret the drawings separately, leading to recurring issues such as terminal connection errors or repeated requests for terminal location changes. In particular, when frequent changes to terminal locations are required due to site conditions, even for the same model, conventional systems are unable to adequately support prior review or the presentation of alternatives.

[0008] Meanwhile, the aforementioned background technology is technical information that the inventor possessed for the derivation of the present invention or acquired during the process of deriving the present invention, and it cannot necessarily be considered publicly known technology disclosed to the general public prior to the filing of the present invention. The problem to be solved

[0010] The present invention has been devised to solve the aforementioned problems and aims to provide a drawing-based automatic specification extraction and verification ordering system that automatically extracts dimensional, structural, and specification information from drawings and specification images of various facilities or devices including lighting fixtures, performs verification and correction by comparing the extracted information with a standard catalog, and automatically generates wiring connection guidance and order forms.

[0011] The technical problems of the present invention are not limited to those mentioned above, and other unmentioned technical problems will be clearly understood by those skilled in the art from the description below. means of solving the problem

[0013] A drawing-based automatic specification extraction and verification ordering system according to one embodiment of the present invention automatically extracts dimension, structural information, and specification information from drawings and specification images, performs verification and correction by comparing the extracted information with a standard catalog, and automatically generates wiring connection guides and order forms.

[0014] The above drawing-based automatic specification extraction and verification ordering system is,

[0015] A data analysis unit that recognizes dimension lines, arrows, labels, and symbols from a drawing image to extract structural information including dimensions, height, hole pattern, and structural information of a lighting fixture, and extracts specification information including material information, luminous flux, color temperature, and power specifications from a table or text area within a specification image;

[0016] A normalization unit configured to compare structural information and specification information extracted by the above data analysis unit with standard catalog information, determine whether the extracted value is suitable by considering the tolerance range set for each manufacturer or product family, and automatically suggest an alternative having the same luminous flux group, the same color temperature group, or equivalent specifications if it deviates from the tolerance;

[0017] A scenario management unit configured to recognize a wiring diagram or wiring symbol within the above drawing image to interpret wiring connection information including terminal location, polarity, and grounding status, automatically generate wiring guidance information based on the interpreted wiring connection information, and provide a terminal location change option for models with frequent requests for terminal location changes; and

[0018] It includes an automatic purchase order generation unit that determines whether mounting brackets or accessories need to be included based on the above structural information, specification information, and wiring connection information, recommends their basic inclusion, and generates a purchase order template with site name, quantity, and delivery date information automatically filled in.

[0019] The normalization unit calculates a normalized value for each of the multiple specification items, including luminous flux, color temperature, power consumption, power specifications, external dimensions, and weight, among the specification information extracted by the data analysis unit, by referring to predefined or learned tolerance ranges and weights for each manufacturer or product family.

[0020] The above item-specific normalization value is calculated as a dimensionless deviation value by dividing the difference between the extracted specification value and the corresponding specification value in the standard catalog by the above tolerance range, and is configured to determine the corresponding specification item as suitable if the above dimensionless deviation value is below a reference threshold, and as unsuitable if it exceeds the reference threshold.

[0021] The above normalization unit is,

[0022] It is configured to calculate a suitability score for each specification item by applying the weights to the normalized values ​​for each item, and to calculate a comprehensive normalized similarity score for each candidate product within the standard catalog by summing or weighting the suitability scores for each specification item.

[0023] Based on the above comprehensive normalized similarity score, a plurality of candidate products within the above standard catalog are rearranged in order of similarity, and candidate products whose comprehensive normalized similarity score is greater than or equal to a threshold value are selected as a suitable candidate group.

[0024] The normalization unit is configured to automatically identify as replacement candidates products in which the luminous flux or color temperature belongs to the same group and the power consumption or power specifications fall within the allowable tolerance range by comparing with specification information extracted from the above suitable candidate group, and to generate replacement reason information for the replacement candidates including the difference items compared to the original specifications, the degree of difference, and whether the allowable tolerance is satisfied.

[0025] The above normalization unit is characterized by being configured to classify a pre-set core specification item's normalization value as a restricted suitable candidate and provide it to the user for selection, even if the above-mentioned comprehensive normalization similarity score is below the reference value, if the normalization value of the pre-set core specification item among the above-mentioned item-specific normalization values ​​is included within the reference range.

[0026] The above data analysis unit is,

[0027] After detecting the dimension line, arrows placed at both ends of the dimension line, and dimension text labels from the uploaded drawing image, a reference length section indicated by the dimension line is specified using the centerline direction of the dimension line and the tip direction of the arrow, and a scale factor is calculated by correlating the dimension value indicated by the dimension text label with the pixel length of the reference length section in the image coordinate system.

[0028] When calculating the above scale factor, using the relative positional relationship of drawing components including the pixel distance between the arrows at both ends of the dimension line, the distance at which the placement position of the dimension text label is spaced in the normal direction of the dimension line, and whether there is an intersection or overlap between the dimension line and an adjacent line segment, a dimension line candidate corresponding to the dimension text label is selected and the dimension line-label matching is confirmed.

[0029] For multiple dimension lines among the dimension lines for which the above dimension line-label matching has been confirmed, multiple scale factor candidates are calculated based on the pixel length of each dimension line and the corresponding dimension value, and a representative scale factor is determined after removing outliers so that the variance of the multiple scale factor candidates becomes less than or equal to a reference value.

[0030] Based on the above representative scale factor, arbitrary structural elements within the drawing image (hole center point, pitch between holes, length of the outer contour, height dimension, or separation distance from the mounting surface) are converted from pixel lengths measured in an image coordinate system into actual measurement units to calculate structural information including the dimensions, height, and hole pattern of the lighting fixture, and

[0031] When the structural information calculated as a result of applying the above representative scale factor deviates from the allowable error range of the standard catalog information referenced by the normalization unit, the data analysis unit estimates the rotational distortion of the drawing image based on the direction vector of the dimension line and the tip coordinates of the arrow, applies a corrective rotational transformation, recalculates the scale factor so that the linearity of the dimension line, the symmetry between arrows, and the alignment state of the dimension text label satisfy the reference conditions, and is configured to re-correct the structural information using the recalculated representative scale factor and output it. Effects of the invention

[0033] According to one aspect of the present invention described above, by integrating the entire process of automatically extracting and verifying structural and specification information of a lighting fixture from images of drawings and specification documents to generate a purchase order into a single processing flow, various problems that occurred in conventional manual ordering methods or simple character recognition-based systems can be effectively resolved. Brief explanation of the drawing

[0035] FIG. 1 is a diagram showing the schematic configuration of a drawing-based automatic specification extraction and verification ordering system in one embodiment of the present invention. Specific details for implementing the invention

[0036] The following detailed description of the invention refers to the accompanying drawings, which illustrate specific embodiments in which the invention may be practiced. These embodiments are described in sufficient detail to enable those skilled in the art to practice the invention. It should be understood that various embodiments of the invention are different but need not be mutually exclusive. For example, specific shapes, structures, and characteristics described herein may be implemented in other embodiments without departing from the spirit and scope of the invention in relation to one embodiment. It should also be understood that the location or arrangement of individual components within each disclosed embodiment may be changed without departing from the spirit and scope of the invention. Accordingly, the following detailed description is not intended to be limiting, and the scope of the invention is limited only by the appended claims, including all equivalents to those claimed therein, provided appropriately described. Similar reference numerals in the drawings refer to the same or similar functions across various aspects.

[0037] In this specification, terms such as “comprising” or “having” are intended to indicate the existence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof. Where one component is described as being “connected” or “combined” with another component, it may be directly connected to or connected to that other component, but there may be other in between.

[0038] It should be understood that components may exist. On the other hand, when it is stated that one component is "directly connected" or "directly coupled" to another component, it should be understood that no other component exists in between.

[0039] Meanwhile, a "module" or "part" for a component as used in this specification performs at least one function or operation. Furthermore, a "module" or "part" may perform a function or operation by hardware, software, or a combination of hardware and software. Additionally, a plurality of "modules" or a plurality of "parts," excluding a "module" or "part" that must be performed on specific hardware or on at least one processor, may be integrated into at least one module. A singular expression includes a plural expression unless the context clearly indicates otherwise.

[0040] In addition, power, power transmission, and control therefor for the following assembly configurations and embodiments, including "by control," follow conventional technology including terminals, applications, hardware control modules, etc., so they are omitted to avoid redundancy.

[0041] In addition, the operation embodiments and configurations described in a general manner without being explained in detail below follow the prior art and are omitted in order to focus on describing the purpose of the present invention and the resulting effects.

[0042] Furthermore, in describing the present invention, if it is determined that a detailed description of related known functions or configurations may unnecessarily obscure the essence of the invention, such detailed description is abbreviated or omitted.

[0043] Hereinafter, preferred embodiments of the present invention will be described in more detail with reference to the drawings.

[0044] FIG. 1 is a diagram showing the schematic configuration of a drawing-based automatic specification extraction and verification ordering system according to one embodiment of the present invention.

[0045] The drawing-based automatic specification extraction and verification ordering system according to the present invention aims to automatically extract dimensional, structural, and specification information from drawings and specification images of various facilities or devices including lighting fixtures, perform verification and correction by comparing the extracted information with a standard catalog, and automatically generate wiring connection guides and order forms.

[0046] To this end, the drawing-based automatic specification extraction and verification ordering system according to the present invention can be implemented in a management server (hereinafter referred to as the server).

[0047] The server may be a self-contained server for the service according to the present invention, a cloud server, or a peer-to-peer (P2P) set of distributed nodes.

[0048] The server can perform one or more of the computation, storage, reference, input / output, and control functions of a general computer, and may include an artificial neural network described below based on input data.

[0049] The server may include a processor and memory. The processor may perform a smart interior service according to the present invention and may include devices capable of performing such service. The processor may execute a program or control the server. Program code executed by the processor may be stored in memory. The memory may store relevant information for performing the service according to the present invention or a program for implementing a method. The memory may be volatile memory or non-volatile memory.

[0050] The server can use a network to send data to an external device or receive data from an external device.

[0051] The server can train artificial neural networks and can also utilize artificial neural networks once training is complete. The processor can train or execute artificial neural networks stored in memory, and memory can store artificial neural networks once training is complete. The electronic device that trains the artificial neural network and the electronic device that uses it may be the same, but they may also be separate.

[0052] The artificial intelligence-related functions according to the present disclosure are operated through a processor and memory. The processor may be composed of one or more processors. In this case, the one or more processors may be general-purpose processors such as CPUs, APs, and DSPs (Digital Signal Processors), graphics-dedicated processors such as GPUs and VPUs (Vision Processing Units), or artificial intelligence-dedicated processors such as NPUs. The one or more processors control the processing of input data according to predefined operation rules or artificial intelligence models stored in memory. Alternatively, if the one or more processors are artificial intelligence-dedicated processors, the artificial intelligence-dedicated processors may be designed with a hardware structure specialized for processing a specific artificial intelligence model.

[0053] The predefined rules of operation or artificial intelligence models are characterized by being created through learning. Here, being created through learning means that a basic artificial intelligence model is trained using a number of training data by a learning algorithm, thereby creating predefined rules of operation or artificial intelligence models configured to perform desired characteristics (or objectives). Such learning may be performed on the device itself where the artificial intelligence according to the present disclosure is executed, or it may be performed through a separate server and / or system. Examples of learning algorithms include supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning, but are not limited to the examples described above.

[0054] An artificial intelligence model can be composed of multiple neural network layers. Each of the multiple neural network layers has multiple nodes and weight values, and

[0055] A neural network through operations between the computation results of the previous layer and multiple weights

[0056] It performs operations. Multiple weights possessed by multiple neural network layers can be optimized based on the learning results of the artificial intelligence model. For example, multiple weights can be updated so that the loss value or cost value obtained from the artificial intelligence model during the learning process is reduced or minimized. Additionally, to minimize the loss value or cost value, multiple weights can be updated in a direction that minimizes the gradient associated with the loss value or cost value. Artificial neural networks may include deep neural networks (DNNs), such as Convolutional Neural Networks (CNNs), Deep Neural Networks (DNNs), Recurrent Neural Networks (RNNs), Restricted Boltzmann Machines (RBMs), Deep Belief Networks (DBNs), Bidirectional Recurrent Deep Neural Networks (BRDNNs), or Deep Q-Networks, but are not limited to the examples mentioned above.

[0057] A network is a network that serves as a transmission path for web pages; it may be a closed network such as a LAN (Local Area Network) or WAN (Wide Area Network), but it is desirable for it to be an open network such as the Internet. The Internet refers to a global open computer network structure that provides the TCP / IP protocol and various services existing at its upper layers, namely HTTP (HyperText Transfer Protocol), Telnet, FTP (File Transfer Protocol), DNS (Domain Name System), SMTP (Simple Mail Transfer Protocol), SNMP (Simple Network Management Protocol), NFS (Network File Service), and NIS (Network Information Service).

[0058] The terminal can be implemented in various forms. For example, the terminal described in this specification may be a mobile terminal such as a smartphone, tablet PC, PDA (Personal Digital Assistant), PMP (Portable Multimedia Player), or MP3 player, as well as a fixed terminal such as a smart TV or desktop computer.

[0059] Specifically, the system according to the present invention includes a data analysis unit, a normalization unit, a scenario management unit, and an automatic purchase order generation unit.

[0060] The analysis unit is configured to automatically extract structural and specification information of a lighting fixture by receiving images of drawings and specifications of an uploaded lighting fixture, and specifically to correct drawing scale distortion and recalculate structural information by utilizing the relative positional relationships between dimension lines, arrows, and dimension text labels included in the drawing images.

[0061] More specifically, the data analysis unit detects line segment-shaped objects from the drawing image and classifies line segments among the detected segments that are longer than a certain length and maintain a certain angle range as dimension line candidates. At this time, the data analysis unit calculates the centerline direction vector of each dimension line candidate, detects triangle or arrowhead-shaped objects located near both ends of the corresponding dimension line candidate as arrow candidates, and determines whether the tip direction vector of the arrow candidate has an opposing relationship with the centerline direction vector to finalize the pair of line segments and arrows constituting the actual dimension line.

[0062] In addition, the data analysis unit analyzes text areas placed within the drawing image to detect dimension text label candidates and evaluates the relative positional relationship between each dimension text label candidate and the confirmed dimension line. At this time, utilizing a drawing creation rule that dimension text labels are generally placed at a predetermined distance in the normal direction of the dimension line, the data analysis unit calculates multiple evaluation indicators including the distance in the normal direction between the dimension text label and the dimension line, parallelism, and whether there is overlap, and determines the dimension line corresponding to each dimension text label based on the evaluation indicators.

[0063] Once the matching between the dimension lines and dimension text labels is confirmed in this manner, the data analysis unit calculates a reference length interval for each dimension line, defined by the pixel distance between arrows, and derives scale factor candidates by correlating the pixel lengths of the reference intervals with the actual dimension values ​​listed on the dimension text labels. The data analysis unit calculates the mean, variance, and median for the scale factor candidates derived from multiple dimension lines, removes scale factor candidates whose variance exceeds the reference value as outliers, and determines a representative scale factor.

[0064] In addition, the data analysis unit applies the above representative scale factor to measure structural elements such as hole center points, pitch between holes, length of the outer contour, height dimensions, and separation distance from the mounting surface in pixel units within the drawing image, and converts them into actual measurement units to calculate structural information in the first step.

[0065] Meanwhile, the data analysis unit determines that rotational distortion or anisotropic scale distortion exists in the drawing image itself if the structural information calculated in the first step exceeds the allowable error range in the comparison result with standard catalog information, and based on the determination, analyzes the direction vector distribution of the dimension line, the symmetry of the arrow pairs, and the alignment state of the dimension text labels to estimate correction transformation parameters for the drawing image.

[0066] Specifically, the data analysis unit determines that rotational distortion exists if multiple dimension line direction vectors are not distributed parallel to each other, and corrects the drawing image by rotating it in the average direction of those direction vectors. Additionally, if the difference between the scale factor candidates calculated from the horizontal and vertical dimension lines exceeds a reference value, it determines that anisotropic scale distortion exists and recalculates independent scale factors for the horizontal and vertical directions, respectively.

[0067] After such correction transformation is applied, the data analysis unit analyzes the relative positional relationships between dimension lines, arrows, and dimension text labels on the corrected drawing image to recalculate the representative scale factor, and by applying the recalculated representative scale factor to re-correct the structural information, it minimizes dimensional errors caused by differences in scan resolution, shooting angles, or image deformation.

[0068] In other words, the data analysis unit recognizes dimension lines, arrows, and dimension text labels from the uploaded drawing image of the lighting fixture to calculate structural information of the lighting fixture.

[0069] Specifically, the data analysis unit detects straight line segments within the drawing image and identifies dimension line candidates based on the segment length, direction, and repetitive arrangement pattern. At this time, the data analysis unit detects triangular or arrowhead-shaped objects placed near both ends of the dimension line candidates as arrow candidates, and determines the dimension line indicating the actual dimension by determining whether the direction of the tip direction vector of each arrow candidate matches the direction of the centerline direction vector of the dimension line candidate.

[0070] In addition, the data analysis unit analyzes the text area within the drawing image to detect dimension text labels and evaluates the relative positional relationship between the corresponding dimension text labels and the confirmed dimension lines. Specifically, it establishes the dimension lines corresponding to each dimension text label based on criteria such as whether the dimension text label is positioned at a predetermined distance in the normal direction of the dimension line, the degree of parallelism with the dimension line, and whether it overlaps with adjacent line segments.

[0071] Once the correspondence between the dimension line and the dimension text label is established in this manner, the data analysis unit calculates a reference length interval defined by the pixel distance between the arrows at both ends of the dimension line, and calculates candidate scale factors for the drawing image by correlating the pixel length of the reference interval with the actual dimension value listed on the dimension text label. This process is repeated for multiple dimension lines to derive multiple candidate scale factors.

[0072] The data analysis unit calculates the mean and variance values ​​for the plurality of scale factor candidates, determines scale factor candidates whose variance values ​​exceed a pre-set reference variance value as outliers and removes them, and then determines the mean or median value of the remaining scale factor candidates as the representative scale factor.

[0073] Subsequently, the data analysis department applies the above representative scale factor to measure structural elements such as the hole center point, hole pitch, length of the outer contour line, height dimension, and separation distance from the mounting surface in pixel units on the image coordinate system, and converts this into structural information in actual measurement units to calculate primary structural information.

[0074] Meanwhile, the data analysis unit determines that rotational distortion or anisotropic scale distortion exists in the drawing image if the above primary structural information exceeds the allowable error range when compared with the structural information listed in the standard catalog. To this end, the data analysis unit compares direction vectors for a plurality of dimension lines and determines that rotational distortion exists if the direction vectors are not parallel to each other, and applies a rotational correction transformation to the drawing image based on the average direction of the direction vectors.

[0075] In addition, if the difference between the scale factor candidates calculated from the horizontal dimension line and the scale factor candidates calculated from the vertical dimension line exceeds a reference difference value even after rotation correction, the data analysis unit determines that anisotropic scale distortion exists in the drawing image and recalculates independent scale factors for the horizontal and vertical directions, respectively.

[0076] The data analysis unit minimizes dimensional errors that may occur due to differences in scan resolution, shooting angles, or image deformation by recalculating structural information using the recalculated scale factor. This recalculated structural information is subsequently used to determine specification compliance in the normalization unit.

[0077] According to this embodiment, unlike conventional technology that simply recognizes dimension text and uses the value, structural information is calculated through the relative positional relationship between drawing components, statistical verification, and determination of geometric distortion, thereby achieving the effect of significantly improving the reliability of drawing-based structural information.

[0078] In this way, the data analysis unit improves the reliability of structural information—unlike simple character recognition-based technology—through the process of dimension extraction and scale recalibration based on the relative positional relationships between drawing components, and plays a role in substantially enhancing the accuracy of specification conformity judgments subsequently performed by the normalization unit.

[0079] The normalization unit receives structural and specification information extracted and corrected by the data analysis unit, determines whether the said structural and specification information is suitable for ordering in light of standard catalog information, and derives alternative solutions based on verification results that consider tolerances. The normalization unit is configured to quantitatively evaluate the suitability of each specification item by reflecting the tolerance characteristics that differ by manufacturer or product group, rather than relying on a comparison method based on simple value matching.

[0080] Specifically, the normalization unit sets multiple specification items, including luminous flux, color temperature, power consumption, power specifications, external dimensions, weight, and protection class, as reference specification items, and for each reference specification item, refers to tolerance ranges and weights that are predefined by manufacturer or product family or learned based on past order history. In this case, the tolerance range can be set in the form of absolute or ratio values, and the weights are set to reflect the importance of the corresponding specification item in the judgment of order conformity.

[0081] For each specification item, the Normalization Unit calculates the difference between the extracted specification value input from the Data Analysis Unit and the corresponding specification value in the standard catalog, and calculates a dimensionless deviation value by dividing the said difference by the allowable error range corresponding to the specification item. If the dimensionless deviation value calculated in this way is less than or equal to a pre-set threshold, the Normalization Unit determines that the specification item is included within the allowable error range; if it exceeds the threshold, it determines that it is outside the allowable error range.

[0082] In addition, the normalization unit calculates a suitability score for each specification item by applying the weights to the dimensionless deviation values, and calculates an overall normalized similarity score for each catalog product by summing or weighting the calculated suitability scores for each specification item. Based on the overall normalized similarity score, the normalization unit rearranges multiple candidate products within the standard catalog in order of similarity and selects products with an overall normalized similarity score equal to or greater than a threshold value as a suitable candidate group.

[0083] Meanwhile, the normalization unit is configured to classify a product as a restricted candidate for suitability even if the overall normalization similarity score is below the threshold value, provided that the dimensionless deviation value of a pre-specified key specification item, such as luminous flux or color temperature, falls within the allowable tolerance range. This allows the ordering manager to review alternative products that are available for selection in the actual work environment.

[0084] In addition, the normalization unit automatically identifies replacement candidates from among the products included in the suitable or restricted suitable candidate group, by comparing them with extracted specification information; these products belong to the same luminous flux or color temperature group, and their power consumption or power specifications fall within the allowable tolerance range. At this time, the normalization unit is configured to calculate for each replacement candidate the specification items that differ from the original specifications, the magnitude of the difference, and whether the tolerance is satisfied, and to provide this information to the user as the reason for replacement.

[0085] Furthermore, the Normalization Unit is configured to reflect as a deduction factor in the Comprehensive Normalization Similarity Score if structural information corrected by the Data Analysis Unit exceeds the tolerance range of structural information in the standard catalog, thereby reflecting the impact of such structural information on the judgment of order suitability. This ensures that discrepancies in dimensions or mounting structures are appropriately reflected in the specification judgment results.

[0086] In this way, the normalization unit performs a role in reducing judgment errors during the ordering stage by enabling the determination of order suitability and the presentation of reasonable alternatives through normalization operations based on tolerance ranges and weights, which are difficult to derive using simple value comparison methods.

[0087] The scenario management unit of the present invention receives wiring connection information and structural information extracted by the data analysis unit, as well as specification verification results from the normalization unit, and is configured to automatically generate and manage work scenarios required during the installation and wiring process of a lighting fixture. The scenario management unit is configured not merely to visually display wiring symbols within the drawings, but to interpret the meaning of the wiring structure and provide guidance information and selection options that can be directly utilized in the installation work.

[0088] Specifically, the scenario management unit identifies the wiring diagram area or wiring symbol area within the drawing image and detects terminal symbols, polarity indicator symbols, ground symbols, and connector symbols included in the corresponding area. At this time, the scenario management unit analyzes the shape, size, orientation, and connection relationship with adjacent wiring of each symbol to convert wiring connection information, including the location coordinates, terminal number, polarity information, and grounding status of each terminal, into structured data.

[0089] Based on the aforementioned wiring connection information, the Scenario Management Department generates a terminal arrangement map corresponding to the installation direction and mounting reference plane of the lighting fixture. The terminal arrangement map is linked with the external structural information of the lighting fixture and is provided in a form where terminal positions are realigned based on the direction in which the operator faces the fixture during actual installation. This enables even operators with limited experience in interpreting drawings to intuitively identify the connection locations.

[0090] In addition, the Scenario Management Department refers to past order or installation history data to calculate the frequency of requests for terminal location changes for a specific model, and if this frequency exceeds a threshold value, the model is classified as a model with frequent terminal location changes. When classified as such, the Scenario Management Department automatically generates alternative terminal layout options along with a default terminal layout map and provides them to the user for selection.

[0091] In addition, the Scenario Management Department refers to the specification verification results of the Normalization Department to determine cases where a change in the wiring method is necessary due to differences in power specifications or power consumption. For example, if there is a difference between single-phase and three-phase power, or between an external ballast and an internal ballast, the Scenario Management Department is configured to select a wiring scenario corresponding to the specification difference or generate additional wiring guidance.

[0092] The scenario management unit automatically generates wiring guidance information for the wiring scenario created as described above, including the wiring sequence, precautions, and essential verification items. The wiring guidance information includes connection targets for each terminal, polarity warnings, whether grounding is mandatory, and pre-power-on inspection items, and this can be provided in the form of text, icons, or step-by-step guidance.

[0093] Furthermore, the scenario management unit refers to structural information input from the data analysis unit to determine cases where wiring accessibility is restricted depending on the mounting method or installation environment of the lighting fixture. For example, in the case of a recessed lighting fixture where the mounting depth is deep or interference with the ceiling structure is expected, the scenario management unit automatically generates a scenario recommending prior wiring or prior wiring.

[0094] In this way, the Scenario Management Department manages various wiring situations that may actually occur at the installation site on a scenario basis through the interpretation of drawing-based wiring symbols, the linkage of structural information, and the reflection of specification verification results; going beyond the simple provision of wiring information, it plays a role in preventing installation errors in advance and improving work efficiency.

[0095] The automatic purchase order generation unit is configured to automatically generate purchase orders necessary for ordering lighting fixtures by receiving structural and specification information extracted by the data analysis unit, specification conformity judgment results verified by the normalization unit, and wiring scenarios and option information generated by the scenario management unit. Rather than simply generating a document that lists specification values, the automatic purchase order generation unit is configured to automatically output a purchase document that reflects verified information and recommendations to prevent omissions or misunderstandings during the ordering process.

[0096] Specifically, the automatic purchase order generation unit confirms the product selected by the user from among the suitable or alternative candidate groups selected by the normalization unit as the product to be ordered, and sets the catalog code, product name, and key specification items corresponding to the product as the basic items of the purchase order. At this time, the automatic purchase order generation unit is configured to include installation-related information, such as external dimensions, mounting method, and hole pattern, in the purchase order items by referring together with structural information extracted and corrected by the data analysis unit.

[0097] The automatic purchase order generation unit is also configured to include wiring connection methods, terminal arrangement options, and wiring precautions based on power specifications within the purchase order by referring to wiring scenario information entered from the scenario management unit. For example, if a terminal location change option is selected, it is reflected in the purchase order in the form of an option code or comment so that the option is clearly identified.

[0098] In addition, the automatic purchase order generation unit determines the necessity of accessories such as mounting brackets, connectors, and fixing bolts based on installation structure information extracted from drawing images by the data analysis unit. At this time, the automatic purchase order generation unit is configured to comprehensively consider the lighting fixture's mounting method, hole pattern, mounting surface conditions, and accessory-related information within the specifications, and if the inclusion of accessories is required, to set them as default items or indicate them to the user as recommended items.

[0099] In addition, the automatic purchase order generation unit automatically populates order-related input items, such as site name, order quantity, and delivery date, to improve the efficiency of ordering operations. At this time, the automatic purchase order generation unit can refer to past order history or information entered from the project management system to automatically suggest item values ​​that are repeatedly used for the same or similar sites, or set them as default values.

[0100] The automatic purchase order generation unit generates a purchase order document by mapping the confirmed purchase information, as described above, to a predefined purchase order template. The purchase order template is configured to include product information, specification summary, structural information, wiring guide summary, inclusion of accessories, selection of options, site information, and delivery date information. The automatic purchase order generation unit is configured to print the purchase order after verifying whether each item has been filled in without omission.

[0101] Furthermore, the automatic purchase order generation unit is configured to temporarily suspend purchase order creation and provide a warning message to the user if specification discrepancies, structural non-conformities, or missing essential components are detected during the generation process. This prevents potential errors that may occur during the ordering stage in advance.

[0102] In this way, the automatic purchase order generation unit plays a role in simultaneously improving the accuracy and efficiency of ordering operations by automatically generating purchase orders that comprehensively reflect information extracted and verified based on drawings, wiring scenarios, and accessory recommendations.

[0103] The present invention provides a technology that goes beyond merely visually displaying wiring diagrams or wiring symbols included in lighting fixture drawings; it interprets the structural meaning of the wiring symbols to reconstruct terminal locations and polarities, and automatically generates terminal location change options based on the installation environment and the frequency of user requests.

[0104] Specifically, the system of the present invention identifies a wiring diagram area within a drawing image and then detects terminal symbols, polarity indicator symbols, ground symbols, and connector symbols included in that area. At this time, each symbol is not treated as a simple graphic object, but is classified as a functional attribute of a terminal based on the symbol's shape, orientation, and connection relationship with adjacent wiring. For example, a circular or square terminal symbol is recognized as a reference point indicating the terminal location, and a plus or minus sign or ground indicator symbol adjacent to the symbol is mapped to an attribute indicating the polarity or grounding status of the corresponding terminal.

[0105] Based on the terminal attribute information recognized in this manner, the system of the present invention reconstructs the terminal arrangement on the drawing based on the actual installation direction by referencing the external structure information and installation reference plane information of the lighting fixture together. That is, rather than maintaining the terminal positions expressed in the drawing as simple coordinates, it generates a terminal arrangement map based on the direction viewed by the operator at the site, assuming the case where the lighting fixture is installed on a ceiling, wall, or recessed structure.

[0106] Furthermore, the system of the present invention analyzes past order or installation history data to calculate the frequency of terminal location change requests for a specific model. For models where terminal location change requests occur at a frequency exceeding a pre-set threshold, the system classifies the model as a model with frequent terminal location changes and automatically generates multiple alternative terminal arrangement scenarios along with a basic terminal arrangement map. At this time, the alternative terminal arrangement scenarios may be generated by rotating the terminal location by a certain angle, changing the terminal order, or moving it to a location with high wiring accessibility.

[0107] In the process of automatically generating such terminal arrangement scenarios, the system of the present invention is configured to derive terminal position change options only within the range where actual manufacturing and installation are feasible by considering the spacing between terminals, external dimensions, the minimum bending radius of the wiring, and the presence of interference with adjacent structures. Through this, it provides the user with only realistic alternative options, rather than merely proposing design changes.

[0108] In addition, the system of the present invention automatically recognizes cases where the wiring method changes due to differences in power specifications or power consumption, and presents corresponding wiring scenarios. For example, even for lighting fixtures with the same external structure, if the wiring sequence and terminal usage methods differ depending on whether a single-phase power supply is used or a three-phase power supply, or whether an external ballast is used or an internal ballast, the system of the present invention selectively provides wiring guidance and terminal arrangement scenarios that reflect the differences in specifications.

[0109] As such, the wiring diagram symbol-based terminal location and polarity reconstruction technology according to the present invention is technically differentiated in that, unlike conventional technology that merely displays or transmits wiring information expressed in a drawing, it automatically generates wiring scenarios and terminal location options optimized for the actual installation environment by comprehensively analyzing drawing information, structural information, and user behavior information.

[0110] As a result, the present invention enables the identification of terminal position mismatch, polarity error, and wiring accessibility issues that frequently occur during the installation of lighting fixtures at the ordering stage and the presentation of alternatives, thereby achieving significant effects such as reduced installation errors, prevention of rework, and improved overall construction efficiency.

[0111] In order to resolve the problem of missing accessories, such as mounting brackets, connectors, etc., that frequently occur during the ordering process for lighting fixtures, the present invention provides a technology that automatically infers whether accessories need to be included by comprehensively analyzing structural information included in drawings and descriptive information listed in specifications, and reflects this as a standard inclusion or recommended item at the ordering stage.

[0112] Specifically, the system of the present invention receives structural information such as the external dimensions, hole pattern, mounting surface shape, mounting direction, and whether it is recessed, extracted by a data analysis unit, and determines the installation context of the lighting fixture based on this. For example, if the external outline and mounting hole pattern of the lighting fixture are shown together in a drawing and no separate fixing structure is shown, the lighting fixture is determined to be a structure that requires an independent mounting bracket or bracket.

[0113] Furthermore, the system of the present invention analyzes text or table information within a specification image and refers to whether accessories are listed. At this time, it recognizes expressions such as include, exclude, option, optional, sold separately, and purchase separately, and determines whether such expressions are linked to an attachment stand or a specific accessory item. Through this, it distinguishes whether an accessory is included by default, is an optional choice, or is explicitly excluded from the specification.

[0114] The system of the present invention combines the result of determining the installation context based on the structural information with the contents of the specification to infer whether the inclusion of accessories is necessary in the form of a probability or a score. For example, if a mounting bracket is determined to be structurally essential and there is no explicit exclusion statement in the specification, the system is configured to set the corresponding mounting bracket as a default included item. Conversely, if it is specified as an optional item in the specification, it is provided to the user as a selectable item, but a recommendation is indicated with a default checked state.

[0115] Furthermore, the system of the present invention refers to past order or installation history data to analyze the frequency of additional orders or on-site issues caused by missing accessories for lighting fixtures having the same or similar structure. If the issue of missing accessories occurs above a standard frequency, the relevant accessory is classified as an automatically recommended accessory and is subsequently treated as a standard inclusion or mandatory verification item when ordering lighting fixtures with the same or similar structure.

[0116] In addition, the system of the present invention refers to the specification verification results of the normalization unit and automatically reflects cases where the required accessory configuration changes due to structural differences between the selected product and the substitute product. For example, even if they are substitute products of the same luminous flux group, if additional brackets are required because the external dimensions or mounting method differ, the system of the present invention automatically includes the corresponding accessory in the order form or clearly informs the user.

[0117] As such, the automatic inference technology regarding the inclusion of mounting brackets and accessories according to the present invention is technically distinct in that it interprets the installation context by combining drawing-based structural information, specification text information, and past ordering history, thereby preventing the possibility of missing accessories during the ordering stage. Unlike conventional technologies that merely provide a list of accessories or allow the user to select them, this technology is recognized as having an inventive step in that it actively determines and recommends the inclusion of accessories by reflecting the actual installation environment and the history of past problems.

[0118] As a result, the present invention provides the effect of significantly reducing reordering, process delays, and cost increases caused by missing parts during the lighting fixture ordering process, and improving completeness at the ordering stage.

[0119] In some other embodiments, the automatic purchase order generation unit aims to prevent reordering and construction delays caused by missing parts or unclear modification requests occurring during the ordering stage. A system for implementing this may be configured to include a data receiving unit (100), a text analysis module (200), a rule-based processing module (300), a terminal location processing module (400), a document generation module (500), a user verification module (600), and a data storage unit (700).

[0120] The data receiving unit (100) is defined as a hardware component that receives raw data related to orders in an unstructured text format from an external system, and serves as the initial data inflow interface of the system. The data receiving unit (100) performs the function of transmitting the received data stream as is to the text analysis module (200) described later, without any prior processing or verification of the format or content of the data received through various communication protocols such as SMTP and API gateways. As a method of transmitting and receiving data, it receives TCP / IP-based data packets from an external ordering system, an email client, or a messenger server, and transmits the body of the received data to the text analysis module (200) via an internal system bus. In one embodiment, when a representative of an external business partner sends an email with the content "Ordering Model AB-200 for a new exhibition hall," the data receiving unit (100) receives the entire body text of the email as raw data and executes a command to immediately transmit it to the text analysis module (200).

[0121] The text analysis module (200) is defined as a software component that evaluates the semantic clarity of received raw data from various angles to quantify the potential risk of an order and identify new patterns that the system must learn. The text analysis module (200) performs a 'order ambiguity grade determination process' and, specifically, sequentially refers to the 'non-normal keyword dictionary', 'model-specific essential specification table', and 'conflict rule set' of the data storage unit (700) to determine whether there are 'non-normal expressions', 'missing essential information', or 'semantic conflicts', and performs the function of determining the final 'order ambiguity grade' by summing the 'risk scores' of each factor. In addition, if a new pattern that cannot be processed by existing rules is detected, it performs the function of recording it as an 'unclassified pattern log' according to the 'rule learning target pattern identification method'. As a method for transmitting and receiving data, raw data is received from the data receiving unit (100), and the determined 'order ambiguity grade' and original order data are transmitted to the rule-based processing module (300) and automatic processing reliability determination logic described later, and the 'unclassified pattern log' and the final processing history therefor are recorded in the data storage unit (700). In the above embodiment, regarding the text "for new exhibition hall," the text analysis module (200) identifies it as a 'non-normal expression' and identifies missing essential accessory information by referring to the 'model-specific essential specification table', calculates a risk score by synthesizing this information, and finally executes a command to determine the 'order ambiguity grade' as 'caution' and transmit it to a subsequent module.

[0122] The above rule-based processing module (300) is defined as a software component that comprehensively analyzes the technical characteristics, installation environment, and historical data of the ordered product to hierarchically determine and recommend the necessity of potential accessories omitted from the order form. The above rule-based processing module (300) performs a 'potential accessory automatic recommendation process' and, specifically, sequentially references the 'model-accessory dependency table', 'site type-accessory requirement table', and 'past order pattern database' of the data storage unit (700). Through this, it calculates the 'necessity score' of each potential accessory to determine the final 'necessity grade' and performs the function of automatically including it in the order details or adding it as a recommended item. As a method of transmitting and receiving data, it receives the 'order ambiguity grade' and order data from the above text analysis module (200), and transmits the 'automatically added / recommended accessory list' and the 'necessity grade', which are the processing results, to the automatic processing reliability determination logic to be described later. Based on the information that it is for a new exhibition hall in the above embodiment, the rule-based processing module (300) refers to the 'site type-accessory requirement table' to determine that the 'ceiling fixing bracket' is 'essential' grade and the 'honeycomb louver' is 'recommended' grade, and executes a command to transmit the processing result including this information to the next step.

[0123] The above-described automatic processing reliability determination logic is defined as a core integrated function of the system that determines the final reliability of the system's automatic processing results by synthesizing the processing results of the text analysis module (200) and the rule-based processing module (300), and accordingly controls subsequent actions of the system differentially. The logic performs an 'automatic processing reliability determination process' and specifically performs the function of determining the final 'automatic processing reliability' grade by referring to a judgment rule matrix with the input 'order ambiguity grade' and 'system intervention level' as two axes. As a method of transmitting and receiving data, the processing results are received from the text analysis module (200) and the rule-based processing module (300), respectively, and the determined 'automatic processing reliability' grade is transmitted to the document generation module (500) and user verification module (600), which will be described later, to control their operations. In the above embodiment, if the 'order ambiguity grade' is evaluated as 'caution' and the 'system intervention level' is evaluated as 'medium', the logic determines the final 'automatic processing reliability' as 'medium' according to the matrix rule and executes a command to transmit this result to the user verification module (600).

[0124] The document generation module (500) is defined as a software component that automatically generates a final production order for orders with secured reliability, and the user verification module (600) is defined as a software component that provides an interface when human review or approval of the system's judgment is required. The operation of the two modules is exclusively controlled by the 'automatic processing reliability' grade. The document generation module (500) is activated to generate a production order only when a 'high reliability' grade is received, and when a 'medium reliability' or 'low reliability' grade is received, the user verification module (600) is activated to perform the 'automatic confirmation after notification' or 'waiting for explicit approval' procedure according to the grade. As a method of transmitting and receiving data, control signals and final order data are received from the automatic processing reliability determination logic. The document generation module (500) transmits the generated production order to an external production management system, and the user verification module (600) receives the user's response and records the final confirmed data in the data storage unit (700). In the above embodiment, since the 'automatic processing reliability' is determined to be 'medium', the user verification module (600) is activated to request the user to verify the relevant content, and if there is no response within 12 hours, a command is sent to automatically confirm the content.

[0125] The data storage unit (700) is defined as a physical storage device that supports the dynamic evolution of the system by storing all rules, data, and history necessary for the judgment and learning of each module and accumulating new information generated during the system operation process. The data storage unit (700) stores static and dynamic data such as 'model-specific essential specification tables', 'collision rule sets', and 'past order pattern databases', and provides the corresponding data according to the request of each module. It also performs an active function of recording 'unclassified pattern logs' and the user's final processing history regarding them according to the 'rule learning target pattern identification method', and generating and managing a 'rule update candidate list' based on this. As a method for transmitting and receiving data, it transmits and receives data bidirectionally with all modules within the system, and in particular, receives the final confirmed processing history from the user verification module (600) and utilizes it as source data for statistical learning. In the above embodiment, when the user's final confirmation history for the keyword "exhibition hall" is repeatedly recorded, the count of 'potential rule candidates' in the data storage unit (700) increases, and when this count exceeds a threshold, a command is executed to automatically create and save a new rule in the 'rule update candidate list'.

[0127] Below, a specific processing process is described in which the above components are organically linked to achieve the purpose of the present invention.

[0128] The overall operation of the system according to the present invention consists of four major core processes.

[0129] The purchase order ambiguity determination process comprehensively identifies potential error factors, such as irregular expressions, missing information, and semantic conflicts, contained in the original received purchase order data. It then quantitatively evaluates the risk each factor may pose to the production process to determine the reliability of the purchase order itself as 'Clear', 'Caution', or 'Ambiguous'.

[0130] The automatic recommendation process for potential accessories hierarchically analyzes the technical characteristics of the ordered lighting fixture model, environmental constraints of the installation site, and past statistical ordering patterns, even if not specified in the purchase order, to determine the necessity of missing accessories as 'Essential', 'Recommended', or 'Reference' grades and automatically reflects this in the order details.

[0131] The automatic processing reliability determination logic cross-evaluates the 'order ambiguity grade' determined in the first process and the 'system intervention level' resulting from the second process in a matrix form, and determines the overall reliability of the final order content automatically supplemented by the system as 'high', 'medium', or 'low' grades.

[0132] The method for identifying pattern targets for rule learning involves the system independently detecting repetitive manual processing tasks, verifying the validity and repeatability of the processing history to select patterns worthy of learning as new automation rules, and then proposing them to the manager to gradually expand the scope of the system's automation.

[0133] The text analysis module (200) according to the present invention performs the function of using a plurality of judgment criteria predefined in the data storage unit (700) to evaluate the semantic clarity of an order document. Each judgment criterion is classified by type of key causes that caused rework, production delays, and customer complaints during past order processing, and through this, it executes a process to identify potential order errors in advance and quantitatively evaluate them.

[0134] The first judgment criterion determines whether non-standard expressions are included. These non-standard expressions are defined as subjective or ambiguous natural language expressions that require additional interpretation by the worker, such as "at your own discretion," "appropriately," and "opposite side," rather than standardized terms that the system can immediately process, such as "left," "right," and "rear." This criterion is configured to prevent manufacturing errors caused by the worker's arbitrary interpretation and to clearly process all order specifications based on data. The second judgment criterion is whether essential information is missing. This is defined to mean a state in which essential parts or essential specification information defined in the 'model-specific essential specification table' within the data storage unit (700) for the ordered lighting fixture model is not explicitly listed in the order document. This criterion performs the function of fundamentally preventing on-site construction delays and additional order costs caused by the omission of parts. The third judgment criterion is whether a semantic conflict occurs, which is defined as a state in which two or more requirements that cannot be technically or physically implemented simultaneously exist together within a single order request. For example, the simultaneous existence of requests for 'rear center terminal location' and 'left side perforation' can cause physical conflicts; this standard is designed to detect critical specification conflicts early during the design and manufacturing phases, thereby reducing unnecessary prototype production and disposal costs.

[0135] The text analysis module (200) is configured to determine the final ambiguity grade by applying the three judgment criteria sequentially and in combination. This performs a sequential procedure of factor identification, risk assessment, and grade comprehensive determination to differentially evaluate the risk that each factor poses to the actual production process, going beyond simply calculating the number of ambiguity factors. As a technical basis, it executes a command to accurately determine the actual risk level of the purchase order by adopting a logical procedure that evaluates the risk differently according to each criterion and combines them, reflecting the empirical knowledge from the industrial field that the omission of essential information has a more fatal impact on the production process than simple non-regular expressions, and that obvious semantic conflicts have a more fatal impact than the omission of information.

[0136] The text analysis module (200) analyzes the text of the order document to determine whether each judgment criterion is satisfied and performs processing to convert it into an internal risk score. In the case of non-regular expressions, it compares them with about 100 words such as 'self-as-you-can', 'appropriately', and 'opposite side' included in the 'non-regular keyword dictionary' of the data storage unit (700) and processes to assign 1 point whenever a matching word is found. In the case of missing essential information, it queries the 'model-specific essential specification table' based on the ordered model and performs calculations such as assigning 5 points if there is no mention of 'SMPS', which is an essential accessory of a specific model, and assigning 3 points if there is no mention of 'color temperature', which is an essential specification. In the case of semantic conflicts, it determines based on the 'conflict rule set' of the data storage unit (700) and executes a command to assign 10 points whenever a conflict rule is detected, for example, 'the terminal location is the rear and the perforation location is the side'.

[0137] The system automatically determines subsequent actions based on the final determined order ambiguity grade. In the 'Clear' grade, i.e., when the total risk score is 0 points and no ambiguity is found, the system immediately transmits the order data to the next step, the rule-based processing module (300), without separate user intervention, and commands the system to continue automatic processing. In the 'Caution' grade, i.e., when minor ambiguity is found with a total risk score between 1 and 9 points but the system has the potential to automatically correct it, a 'Verification Required' notification is sent to the order manager via the user verification module (600); however, if no separate action is taken within a preset time of 12 hours, the system automatically sends the notification to proceed with processing at default values. In the 'Ambiguous' grade, i.e., when the total risk score is 10 points or higher and there is a possibility of a fatal error, the order processing process is immediately stopped, and 'Urgent Verification and Correction' is requested via the user verification module (600); the system controls the system so that it does not proceed to the next step without explicit correction and approval from the manager.

[0138] As a specific scenario, a processing procedure is performed step by step when an order form containing the following content is received through the data receiving unit (100): "Ordering 50 units of Model DS-100A. Please position the terminals on the rear, and drill a suitable wire entry point on the left." In the first step, the text analysis module (200) begins analyzing the order form text. In the second step, it is discovered that the keyword "suitably" matches the "irregular keyword dictionary," and a risk score of 1 is assigned. In the third step, by referring to the "model-specific essential specification table," it is confirmed that the essential items for the model, "SMPS" and "color temperature," are both missing, and additional risk scores of 5 and 3 are assigned, respectively. In the fourth step, the two phrases "terminal position on the rear" and "wire entry point on the left" are identified, and it is detected that they correspond to the relevant rules of the "conflict rule set," and an additional risk score of 10 is assigned. In the fifth step, the total risk scores are arithmetically summed to calculate 1 + 5 + 3 + 10 = 19 points. In the sixth step, since the sum of 19 points satisfies the criterion of 10 points or more for the 'ambiguous' grade, the final 'order ambiguity grade' is determined as 'ambiguous'. In the final seventh step, the system stops order processing and sends an urgent verification request to the person in charge via the user verification module (600) due to semantic conflict and missing essential information.

[0139] The judgment logic of the present invention includes measures to minimize the possibility of false positives or missed detections, and is configured to record the details of final modifications and confirmed contents made by the person in charge for all orders judged to be 'ambiguous' or 'caution' grades in the data storage unit (700) as log data. The system performs the function of continuously improving the accuracy of the judgment logic over time by periodically analyzing the log data to automatically update the 'non-regular keyword dictionary', 'conflict rule set', etc., or by recommending updates to the administrator.

[0140] When comparing the prior art, a simple keyword-based system, with the present invention, when the content "Model DS-100A, rear of terminal, left side appropriately perforated" is entered, the prior art ignores the expression "appropriately" and fails to analyze the relationship between the two requests or the omission of essential information, thereby failing to recognize errors and processing the request, which causes production delays. In contrast, the present invention identifies irregular expressions, omission of essential information, and semantic conflicts, comprehensively assesses the risk level to determine it as "ambiguous," immediately stops processing at the order acceptance stage, and sends a request for correction to the person in charge. As a result, the present invention achieves the effect of preventing rework and delivery delays by blocking potential errors in advance through multi-faceted analysis and clarifying processing priorities.

[0141] The judgment rule is set so that if the total risk score is 0, the final grade is determined as 'Clear' and automatic processing is performed immediately; if the score is between 1 and 9, the grade is determined as 'Caution' and the person in charge is notified and processed automatically after a period of time; and if the score is 10 or higher, the grade is determined as 'Ambiguous' and processing is stopped and waiting for approval from the person in charge. According to ten different simulation examples, consistent judgment results are output, such as DS-100A, rear terminal, including SMPS, 5000K, where all conditions are clear, the grade is assigned as 'Clear' with a total score of 0; DS-100A, with multiple ambiguous factors, the grade is assigned as 'Ambiguous' with a total score of 10; and DS-200B, with only essential information missing, the grade is assigned as 'Caution' with a total score of 5.

[0142] The judgment rules and risk scoring system presented in this invention are not arbitrarily set, but are configured based on the results of statistically analyzing the frequency and severity of major causes that led to actual production errors and customer claims by analyzing thousands of order and service data accumulated over the past several years. In other words, the highest risk score of 10 points was assigned to 'semantic conflict' because it caused the highest cost loss when such an error occurred; therefore, the judgment rules of this invention serve as an objective and inevitable technical means to minimize cost losses due to errors and maximize the efficiency of order processing.

[0143] The adoption of each component yields specific technical effects. By adopting non-standard representation criteria, commands are executed to ensure the uniformity of production quality while eliminating subjective judgment by workers. By adopting criteria for missing essential information, omitted details are mandatorily supplemented during the initial ordering stage, thereby reducing administrative and time costs associated with additional orders. Furthermore, by adopting semantic conflict criteria, physical and technical contradictions are eliminated prior to the design and production phases, thereby performing controls that fundamentally prevent material loss and rework costs. Through a judgment process that organically combines these multiple criteria, technical effects are achieved that significantly increase the detection rate of ordering errors and shorten the overall production lead time compared to conventional manual inspection methods.

[0144] The criteria for terms and threshold values ​​used in this invention are clearly defined. The 'Non-standard Keyword Dictionary' is a database that can be registered, modified, or deleted by the system administrator and is managed to maintain accuracy by continuously adding neologisms or site-specific special terms. The 'Model-Specific Essential Specification Table' is operated in a manner where, whenever a new lighting fixture model is released, the administrator defines essential items based on the technical specifications of that model and registers them in the system. The 'Collision Rule Set' is configured so that lighting fixture design experts define combinations of conditions with potential for physical or electrical interference in the form of logical rules and register them in the system. Finally, the threshold value is statistically set as an initial value through the analysis of past data so that approximately 15% of total orders are classified as 'Caution' and approximately 5% as 'Ambiguous'; this value provides the function for the administrator to monitor and adjust the system's processing efficiency and accuracy as system operation data accumulates.

[0145] The risk score thresholds for determining the above 'Caution' and 'Ambiguous' grades are set by simulating 10 representative order cases from past data that required 'manual processing'. The purpose of the simulation is to find the optimal boundary value that effectively separates cases requiring minor modifications (e.g., simple verification) from cases involving critical errors (e.g., need for redesign).

[0146] [Table 1-1]

[0147]

[0148] As a result of the simulation above, cases requiring simple verification and modification (cases 1 to 6) tend to be distributed at 8 points or less, while cases requiring serious intervention such as design changes or rework (cases 7 to 10) tend to be distributed at 10 points or more. Accordingly, the present invention is characterized by effectively diverting the processing method according to the severity of the problem by setting 10 points, which is the clearest boundary separating these two groups, as the threshold for the 'ambiguous' grade, and classifying scores below that as the 'caution' grade.

[0149] To verify the effectiveness of the set thresholds (Caution: 1–9 points, Ambiguous: 10 points or more), a simulation was performed on 100 random order data records (including 85 normal cases, 10 minor errors, and 5 serious errors). This simulation evaluates how accurately the system detects actual errors (True Positive) and avoids unnecessarily judging normal cases as errors (False Positive).

[0150] [Table 1-2]

[0151]

[0152] As a result of the verification simulation above, the system accurately classified 98 out of a total of 100 cases (Normal -> Clear, Minor -> Caution, Severe -> Ambiguous). In particular, it detected 100% of the 5 severe errors that could cause rework as 'Ambiguous' and demonstrated a low false positive rate by classifying only 1 out of 85 normal cases as an unnecessary 'Caution'. This proves that the set threshold is an effective standard that filters errors that may occur in the field with high reliability while minimizing work inefficiency caused by excessive warnings. This accuracy is calculated using the following conceptual formula.

[0153] Accuracy (%) = [Number of correctly classified cases (diagonal sum)] / [Total number of cases] × 100

[0154] = (84 + 9 + 5) / 100 Х 100 = 98%

[0156] The rule-based processing module (300) according to the present invention performs the function of using a plurality of judgment criteria predefined in the data storage unit (700) to identify and recommend in advance accessories that are essential or recommended in the actual installation and operation environment, even though they are not explicitly stated in the ordering document. Each judgment criterion comprehensively reflects the technical characteristics of the product, physical constraints of the installation environment, and statistical trends derived from past data, thereby performing processing to prevent on-site work delays and additional orders caused by the omission of accessories.

[0157] The first criterion determines the technical correlation between the model and its accessories, which is defined as a technical dependency where normal functionality or physical installation is impossible without specific accessories according to the design specifications of a particular lighting fixture model. For example, a recessed lighting fixture model must necessarily be accompanied by a 'recessed clip' for fixing; this criterion is structured to prevent fundamental ordering errors that fail to meet the product's most basic operating and installation requirements. The second criterion is the site environment-accessory requirement, which is defined as cases where specific accessories are required for product durability, stability, or regulatory compliance based on the characteristics of the installation site specified in the purchase order. For instance, this includes cases where lighting fixtures installed in high-humidity environments require 'IP67 waterproof connectors,' and where 'surge protectors' are strongly recommended when used for 'exterior landscape lighting.' These criteria serve to prevent shortened product lifespans, safety accidents, or performance degradation that may result from general ordering that fails to consider the specific characteristics of the installation environment. The third judgment criterion is a statistical recommendation based on historical data, defined by the statistical trend in which specific accessories were ordered with high frequency when orders under identical or similar conditions were processed in the past. For example, this applies to historical data showing that specific accessories were ordered together in more than 85% of cases when a specific model was installed at a specific site. Although this criterion is not an explicit technical specification or environmental requirement, it implements processing to ensure that accessories customarily used to enhance customer satisfaction or ease of use in actual usage environments are not omitted.

[0158] The rule-based processing module (300) is configured to determine the final necessity grade by sequentially applying the three judgment criteria according to priority. This performs a sequential procedure of a technical correlation determination step, a field environment requirement determination step, a statistical recommendation determination step, and a final necessity grade determination step in order to secure the logical validity and reliability of the recommendation by considering absolute technical necessity as the top priority and then sequentially evaluating environmental factors and statistical trends. As the technical basis, the necessity of accessories has a hierarchical structure of 'indispensable', 'conditionally essential', and 'recommended', and the processing procedure of the present invention reflects this hierarchical structure as is to execute a command to prioritize the identification of technically essential accessories, accessories that become essential under specific conditions, and finally recommended accessories that enhance convenience. This hierarchical approach prevents excessive or unnecessary recommendations and processes to provide the user with a clear basis for judgment.

[0159] The rule-based processing module (300) performs processing to calculate an internal necessity score for each potential accessory by querying each database based on order information. In the case of technical association, it queries the 'model-accessory dependency table' of the data storage unit (700) and processes to assign 100 points to the accessory if the accessory mapped to the 'essential' relationship is missing from the order. In the case of site environment requirements, it analyzes the 'site name' of the order to extract 'site type keywords' and queries the 'site type-accessory requirement table', and performs an operation to assign 50 points if the accessory at the 'essential' requirement level is missing and assign 20 points if the accessory at the 'recommended' requirement level is missing. In the case of statistical recommendation based on past data, it queries the 'past order pattern database' by combining the 'model' and 'site type' of the current order, and executes a command to assign a score based on the value obtained by dividing the calculated simultaneous order probability (%) by 10.

[0160] The necessity grade is determined based on the final necessity score summed for each potential accessory, and the system's recommendation method changes accordingly. For the 'Essential' grade, that is, for technically essential accessories with a total score of 100 points or more, the corresponding accessory is 'automatically included' in the order details, and the fact is 'notified with emphasis' through the user confirmation module (600), and the user is restricted from arbitrarily deleting it. For the 'Recommended' grade, that is, accessories strongly recommended in specific environments or conditions with a total score between 20 and 99 points, the corresponding accessory is added to the order details as a 'recommended item' and presented to the user along with the reason, and the user is configured to selectively add or ignore it. For the 'Reference' grade, that is, optional accessories recommended based on past data with a total score between 5 and 19 points, information is provided in the form of 'products purchased together' in a separate area of ​​the order confirmation screen so that the user can refer to it.

[0161] As a specific scenario, the processing procedure for the case where a purchase order is received for 100 Model DL-500 units to be used as exterior landscape lighting for a coastal hotel is performed step by step. In the first step, a rule-based processing module (300) selects all accessories related to the model as candidates. In the second step, the 'Model-Accessory Dependency Table' is queried to confirm that 'Clip-A' is 'required' for the model, and since it is omitted from the purchase order, 100 points are assigned to 'Clip-A'. In the third step, keywords 'coastal' and 'external' are extracted, and the 'Site Type-Accessory Requirement Table' is queried to confirm that 'Waterproof Connector-IP67' is 'required' for the 'coastal' environment and 'SPD' is 'recommended' for the 'external' environment, and 50 points are assigned to 'Waterproof Connector-IP67' and 20 points to 'SPD', respectively. In the fourth step, assuming there are no significant statistical patterns found in the results of the past data lookup, no points are added. In the fifth step, the final scores are summed up to determine that 'Clip-A' is rated 'Essential' with 100 points, 'Waterproof Connector-IP67' is rated 'Recommended' with 50 points, and 'SPD' is rated 'Recommended' with 20 points. In the final sixth step, the system automatically includes 'Clip-A' in the order details, adds 'Waterproof Connector-IP67' and 'SPD' as recommended items, and sends a command to the user requesting confirmation.

[0162] The logic of the present invention is configured to continuously accumulate the history of a user adopting or rejecting a system’s ‘recommended’ or ‘reference’ suggestion as ‘user feedback data’ in a data storage unit (700). A rule-based processing module (300) analyzes this feedback data to correct the statistical recommendation accuracy of the ‘past order pattern database’ and, for example, performs the function of reducing unnecessary recommendations by automatically lowering the recommendation score weight of a specific accessory when it is continuously ‘recommended’ but rejected by the user more than 90% of the time.

[0163] When comparing the present invention with prior art manual ordering or simple ERP systems, when the same input is given as "Model DL-500, for exterior lighting of a coastal hotel," the prior art causes on-site problems or premature failures because the ordering manager relies on personal experience to manually add accessories, overlooking the importance of "waterproof connectors" and placing the order with them omitted. In contrast, the present invention sequentially and automatically analyzes technical, environmental, and statistical criteria to identify the necessity of "clips," "waterproof connectors," and "SPDs" by grade, thereby guiding the completion of the order with a complete configuration. As a result, the present invention eliminates quality deviations and potential risks based on the knowledge level of the person in charge, and systematizes expert knowledge to enable anyone to place standardized, high-quality orders, thereby achieving the effect of maximizing product durability and safety.

[0164] The judgment rule is set so that if the final necessity score is 100 points or higher, it is judged as 'Essential' and automatically included in the order details; if it is between 20 and 99 points, it is judged as 'Recommended' and added as a recommended item; if it is between 5 and 19 points, it is judged as 'Reference' and information is provided in a separate area; and if it is 4 points or lower, it is judged as 'Low' and no separate action is taken.

[0165] The judgment rules and scoring system of the present invention are not arbitrary decisions but are configured to be based on the product's technical datasheet, domestic and international electrical installation technical standards, and field data accumulated over several years of installation and service processes. 'Technical relevance' is based on product design drawings, 'field environment requirements' are designed to comply with industrial safety standards and product warranty conditions, and 'historical data-based recommendations' utilize only patterns verified for statistical significance from a large volume of actual transaction data. Therefore, the recommendation logic of the present invention serves as the most objective and essential procedure to guarantee product performance and safety and to prevent potential defects.

[0166] The adoption of each component entails specific technical effects. By adopting technical correlation criteria, commands are executed to fundamentally prevent basic product malfunctions and ensure the completeness of orders. By adopting site environment requirements criteria, the specific characteristics of the installation environment are automatically reflected in orders, thereby guaranteeing the product's expected lifespan and significantly reducing failure rates caused by environmental factors. Furthermore, by adopting statistical recommendation criteria, controls are implemented to enhance customer satisfaction by proposing convenience features or performance improvement options preferred by actual users—even if they are not codified standards—without omission. By applying these multiple criteria hierarchically, support is provided for even non-experts to generate orders with expert-level accuracy and completeness, thereby achieving the technical effect of standardizing overall order quality and reducing service costs.

[0167] The terms and threshold criteria used in this invention are clearly defined. The 'Model-Accessory Dependency Table' is created by the product development team based on technical specifications and registered in the system when launching a new product; the 'Field Type-Accessory Requirement Table' is periodically updated by the technical support and quality assurance teams through the analysis of relevant regulations, industry standards, and field claim data; and the 'Historical Order Pattern Database' is operated so that the system automatically anonymizes and stores all order history, and a statistical analysis algorithm is executed on a monthly basis to update meaningful patterns. Finally, thresholds are set as initial values ​​through review by a group of technical support experts and provide a function that allows the system to learn the appropriateness of recommendations and fine-tune them autonomously through 'user feedback data' accumulated over time.

[0169] The necessity score threshold for determining the above 'recommended' and 'reference' grades is set by simulating 10 representative cases in which experts previously manually added accessories. The purpose of the simulation is to find the optimal boundary value that distinguishes between recommendations that directly affect installation quality and stability and optional recommendations that enhance customer convenience.

[0170] [Table 2-1]

[0171]

[0172] As a result of the simulation above, cases judged by experts as 'essential' (cases 8–10) are all distributed at 100 points or higher, cases added for 'stability and performance' (cases 4–7) are distributed at 20 points or higher and less than 100 points, and cases added for 'convenience' (cases 1–3) are distributed at less than 20 points. Accordingly, the present invention is characterized by setting 20 points as the lower threshold for the 'recommended' grade and 100 points as the lower threshold for the 'essential' grade, thereby clearly hierarchizing the importance of recommendations into 'essential', 'recommended', and 'reference' according to objective criteria.

[0173] To verify the effect of the set threshold, the match rate between the components recommended by the system and those added by actual experts was compared for 100 orders with actual shipment history. This simulation evaluates whether the system does not omit essential components (Recall) and whether it does not recommend unnecessary components (Precision).

[0174] [Table 2-2]

[0175]

[0176] As a result of the verification simulation above, the system accurately recommended 20 out of the 21 essential / recommended accessories actually added by experts, demonstrating a high recall rate of 95.2%. This implies that the system almost perfectly prevents the omission of critical accessories. Furthermore, 20 out of the 24 accessories recommended by the system turned out to be actually necessary, recording an accuracy of 83.3%. This proves that the set threshold operates very similarly to the expert's judgment criteria and serves as an effective standard for providing highly reliable recommendations to users. These metrics are calculated using the following conceptual formula.

[0177] Recall (%) = [TP] / ([TP] + [FN]) Х 100

[0178] = 20 / (20 + 1) Х 100 = 95.2%

[0179] Precision (%) = [TP] / ([TP] + [FP]) × 100

[0180] = 20 / (20 + 4) Х 100 = 83.3%

[0182] The system according to the present invention performs a function using two key judgment criteria to evaluate the reliability of the final order content, which the system automatically supplements and confirms without human intervention, by synthesizing the processing results of the text analysis module (200) and the rule-based processing module (300). The purpose of this logic is to execute a process that simultaneously optimizes the speed and accuracy of the entire order processing by balancing the efficiency of complete automation with the stability of manual inspection.

[0183] The first judgment criterion is the clarity grade of the original purchase order, which is the result derived through the aforementioned Process 1 (Purchase Ambiguity Grade Determination Process). It is defined as 'Clear,' 'Caution,' or 'Ambiguous,' indicating how clear and complete the original purchase order received from an external source is. This criterion serves as the most fundamental basis for reliability judgment and is intended to evaluate the quality level of the input data itself. It is structured to reflect the logical premise that if the reliability of the input data is low, the reliability of the subsequent processing result cannot be evaluated as high either. The second judgment criterion is the level of system intervention, which is defined as 'Low,' 'Medium,' or 'High' levels. This indicates how much and to what extent the system added or modified content not present in the original purchase order as a result of executing the aforementioned Process 2 (Automatic Recommendation Process for Potential Accessories). Considering that the risk of potential misjudgment increases as the system's automatic correction actions increase, this criterion performs the function of reflecting the degree of system reasoning and intervention in the reliability evaluation.

[0184] The above-described system is configured to use a matrix-based comprehensive judgment logic to determine the final reliability by combining the two aforementioned judgment criteria. To derive consistent and logical conclusions for various scenarios by cross-evaluating two independent evaluation axes—text clarity and the level of system intervention—it performs a sequential procedure consisting of a text clarity grade verification step, a system intervention level evaluation step, and a matrix rule-based final reliability determination step. As the technical basis, reliable automation must simultaneously satisfy two conditions: 'clear instruction' and 'minimal inference.' The logic of the present invention is designed to determine the final reliability using logic similar to an AND combination method after independently evaluating each of these two conditions. In other words, if the reliability of either of the two conditions is low, the final reliability cannot be high; this represents the most conservative and stable approach from a risk management perspective and possesses a technical necessity to minimize the possibility of errors caused by automation.

[0185] The above system performs a process to calculate an internal intervention score to quantitatively evaluate the 'system intervention level' prior to determining the final reliability. The clarity grade of the original order document receives 'clear', 'caution', and 'ambiguous' as the result of Process 1 above. The system intervention level calculates an intervention score based on the result of Process 2 above, performing an operation in which 5 points are assigned per item when the system automatically adds an accessory of the 'essential' grade, and 10 points are assigned per item when it automatically adds an accessory of the 'recommended' grade. The calculated intervention scores are summed to determine the final 'system intervention level,' and it is judged as 'low' if the score is 0, 'medium' if it is between 1 and 15, and 'high' if it is 16 or higher.

[0186] Depending on the final determined 'automatic processing reliability' grade, the system's follow-up actions are clearly branched into three scenarios. In the 'High' grade, where the reliability of the system's judgment is very high, data is immediately transmitted to the document creation module (500) without a separate user verification procedure to execute a command to automatically generate and issue a production order. In the 'Medium' grade, where the system has performed useful corrections but user review is recommended, the user is notified of the details of the corrections made by the system through the user verification module (600), and if there is no response for 12 hours, which is a preset time, a command to automatically confirm the order is transmitted. In the 'Low' grade, where the original order is highly ambiguous or the system's intervention is excessive and human judgment is essential, all automatic processing procedures are stopped, and 'essential verification and approval' is requested from the person in charge through the user verification module (600).

[0187] As a specific scenario, the processing procedure is executed step by step in the case where a purchase order for Model AB-200, for a new exhibition hall, is received and has gone through Processes 1 and 2, that is, where the result of Process 1 determines the 'Purchase Ambiguity Grade' to be 'Caution' and the result of Process 2 automatically recommends one accessory of the 'Essential' grade and one accessory of the 'Recommended' grade. In the first step, this logic receives the 'Caution' grade resulting from Process 1 as input. In the second step, based on the result of Process 2, the 'Intervention Score' is calculated by adding (1 'Essential' item multiplied by 5 points) and (1 'Recommended' item multiplied by 10 points) to yield 15 points. In the third step, since the total score of 15 corresponds to the criteria for a 'Medium' level, the 'System Intervention Level' is determined to be 'Medium'. In the fourth step, the final reliability is queried by referring to the 'Judgment Rule Table' when the 'Original Text Clarity Grade' is 'Caution' and the 'System Intervention Level' is 'Medium'. In the fifth step, the 'automatic processing reliability' is determined to be 'medium' according to the table rules. In the final sixth step, the system notifies the person in charge through the user verification module (600) that the relevant accessories have been automatically added, and if there is no response within 12 hours, sends a command to initiate a procedure to automatically confirm.

[0188] The reliability determination logic of the present invention is configured to be continuously improved through operational data. The system records the user's final modification and confirmation data for cases judged as 'Medium' or 'Low' grades. If a specific type of order is repeatedly judged as 'Medium' grade and the rate at which users confirm orders without significant modification is very high, the system proposes to the manager to lower the 'intervention score' weight for that type. Through this, it performs a dynamic optimization function that gradually mitigates overly conservative judgments and enhances automation efficiency.

[0189] When comparing the present invention with the prior art batch manual inspection system, when a perfect order form (A) and a partially missing order form (B) are input, the prior art transmits all order forms to the person in charge to verify them one by one, thereby showing inefficiency in that the processing time for the perfect order form (A) is unnecessarily consumed. In contrast, the present invention controls the perfect order form (A) by determining it as 'high reliability' and automatically passing it, while determining the partially missing order form (B) as 'medium reliability' and notifying the user to process it automatically. As a result, the present invention achieves the effect of efficiently allocating resources and maximizing overall processing quality and throughput by automating simple and clear tasks and inducing human intervention only for complex and uncertain tasks.

[0190] The judgment rule table is structured in a matrix form, and when the 'Text Clarity Grade' is 'Clear', reliability is determined as 'High' if the 'System Intervention Level' is 'Low', and as 'Medium' if it is 'Medium' or 'High'. When the 'Text Clarity Grade' is 'Caution', reliability is determined as 'Medium' if the 'System Intervention Level' is 'Low' or 'Medium', and as 'Low' if it is 'High'. When the 'Text Clarity Grade' is 'Absurd', reliability is determined as 'Low' regardless of the 'System Intervention Level'.

[0191] The above judgment rule matrix is ​​not an arbitrary combination but is configured to be based on the concept of a 'quality gate' widely used in manufacturing process control. The original purchase order represents 'input quality,' and the system's automatic correction result represents 'process quality.' The logic of the present invention applies the same principle as cross-verifying these two quality indicators to determine whether to ship the final product. Since cases where the 'original document is ambiguous' or 'system intervention is excessive' indicate a high probability of potential defects, withholding shipment and requiring a detailed inspection serves as an objective and inevitable procedure in terms of quality control.

[0192] The adoption of each component yields specific technical effects. By adopting the text clarity criterion, the quality of input data—the starting point of order processing—is evaluated to execute commands that prevent the fundamental problem of "garbage in, garbage out." By adopting the system intervention level criterion, a mechanism is introduced to monitor and evaluate the automation activities themselves, thereby controlling potential malfunction risks and ensuring reliability. By adopting matrix-based comprehensive judgment rules, the two criteria are organically combined to intelligently distribute the workflow, allowing simple and repetitive tasks to be processed automatically and complex and exceptional tasks to be handled manually with stability. This achieves the technical effect of simultaneously improving the processing speed and stability of the entire system.

[0193] The terms and threshold criteria used in this invention are clearly defined. The 'system intervention level' is an internal evaluation indicator representing how much content the system inferred and modified compared to the original purchase order, and the 'intervention score' is an internal calculation value for quantifying the system intervention level, assigned differentially depending on the type of intervention. Finally, the threshold is statistically set as an initial value by simulating purchase data from the past year so that approximately 70% of all purchase orders are classified into reliability grades of 'High', 20% into 'Medium', and 10% into 'Low'. This value reflects the balance between efficiency and stability and provides a function for the administrator to adjust it according to the system's operational policy.

[0194] The intervention score threshold for determining the above 'system intervention level' is set by simulating 10 past representative cases that required review by a person in charge after automatic correction. The purpose of the simulation is to find the optimal boundary value that distinguishes between minor system intervention (with almost no reduction in reliability) and significant system intervention that absolutely requires review by a person in charge.

[0195] [Table 3-1]

[0196]

[0197] As a result of the simulation above, cases confirmed by the person in charge without any modifications (cases 1 to 4) are all distributed at 15 points or less, while cases that required at least partial modification or a complete re-examination (cases 5 to 10) mostly tend to exceed 15 points (case 6 is on the boundary). Therefore, the present invention is characterized by effectively identifying points where system intervention is likely to require additional judgment from the user by setting 16 points as the lower threshold of the 'high' level and defining 1 to 15 points as the 'medium' level.

[0198] To verify the effectiveness of the established reliability grade classification logic, a simulation was performed on 100 random order data points (70 that can be processed automatically, 20 that require review, and 10 that require manual processing). This simulation evaluates how efficiently the system distributes tasks (automation rate) and how well it prevents errors in automatically processing dangerous orders (stability).

[0199] [Table 3-2]

[0200]

[0201] As a result of the verification simulation above, the system determined 68 out of 70 cases capable of automatic processing out of a total of 100 as 'High Reliability,' achieving a high automation rate of 68%. This means that personnel are freed from unnecessary repetitive inspection tasks. At the same time, not a single one of the 10 risky orders requiring manual processing was incorrectly judged as 'High Reliability,' demonstrating a 100% Risk Prevention rate. This proves that the established logic serves as a standard for effectively achieving the two goals of business efficiency and system stability. These indicators are calculated using the following conceptual formula.

[0202] Automation Rate (%) = [Actual number of cases capable of automatic processing judged as 'High Reliability'] / [Total number of cases] × 100

[0203] = 68 / 100 X 100 = 68%

[0204] Risk Prevention Rate (%) = [Number of cases judged as 'Medium / Low Reliability' out of the actual number of cases requiring manual processing] / [Actual number of cases requiring manual processing] × 100

[0205] = (3 + 7) / 10 X 100 = 100%

[0207] The system according to the present invention performs a function using three key judgment criteria to determine the need to update existing rules stored in a data storage unit (700) in order to actively respond to customer requirements or new types of ordering patterns that change over time and to gradually expand the scope of automation of the system. The purpose of the method is to execute a process that minimizes maintenance costs and secures the self-improvement capability of the system by having the system detect repetitive manual processing tasks on its own and convert them into new automation rules.

[0208] The first judgment criterion is whether a new order pattern is identified, which is defined as a state in which the text analysis module (200) or the rule-based processing module (300) detects a new type of text phrase or request combination that cannot be automatically processed because it does not correspond to any standard rule or keyword predefined in the data storage unit (700) during the processing process. For example, this applies when a request such as manufacturing a special model with 10% improved heat dissipation performance is received for the first time, and this criterion is configured to take the identification of an exception case outside the system's current knowledge range as a starting point. The second judgment criterion is the consistency and validity of manual processing, which is defined as a history in which a person in charge manually intervenes with the identified new order pattern, processes it in a specific and consistent manner, and confirms it as 'normally completed'. If the person in charge processes the same new pattern in a different way each time or ultimately terminates it as 'error' or 'unprocessable', this criterion is not satisfied, and this criterion performs a quality assurance filter function to prevent incorrect or inconsistent manual processing from being learned as a rule, reflecting the principle that the rule to be learned by the system must be a verified and standardized solution. The third criterion is the repeatability of pattern occurrence, defined as a state in which valid manual processing history for a specific type of new order pattern is not a one-time event but accumulates repeatedly more than a threshold number of times within a pre-set specific period. For example, this applies when the same manual processing history for a 'special heat dissipation model' request is recorded five or more times within the last month. This criterion distinguishes between accidental exceptions and ongoing requirements that are becoming new standards, and executes processing to prevent rule overfitting and maintain system stability by targeting only frequently occurring patterns for learning.

[0209] The above system is configured to ultimately select a rule learning target by applying the three aforementioned judgment criteria in a sequential gate manner. As a filtering structure in which progress to the next step is possible only after passing each stage, it performs a sequential procedure consisting of a new pattern identification stage, a manual processing validity verification stage, a repeatability verification stage, and a final learning candidate selection stage to verify the quality and validity of the patterns to be learned step by step. As a technical basis, since adding a new rule in a machine learning system is a significant change that can affect the overall operation of the system, the procedure of the present invention adopts a prudent approach of 'discovery -> verification -> quantification.' It follows a logical flow of first discovering a new phenomenon, qualitatively verifying whether it possesses a valid solution, and finally quantitatively verifying whether it is worthy of generalization; this possesses a technical necessity that guarantees the gradual and stable evolution of the system.

[0210] The above system performs the process of obtaining necessary data from the system log and history database for each step. In the case of identifying new order patterns, the text analysis module (200) automatically records text phrases that it has determined to be unprocessable in the 'unclassified pattern log'. In the case of consistency and validity of manual processing, after the person in charge processes the case, they check the 'final confirmed data' and 'processing completion status' stored in the system. Validity verification is considered passed only if the processing completion status is 'normally completed', and multiple 'final confirmed data' accumulated for the same 'unclassified pattern' are compared, and only if a similarity of 80% or more is shown through a text similarity analysis algorithm is it determined to be 'consistent'. In the case of repeatability of pattern occurrence, the system records the occurrence timestamp of a 'potential rule candidate' in which an 'unclassified pattern' and a 'valid processing history' are paired, and performs an operation to count them.

[0211] The 'rule learning target pattern' finally selected by sequentially passing all three of the above judgment criteria is utilized by the system. First, a command is executed to automatically register the selected pattern and the verified processing method for it in the 'rule update candidate list' database. Whenever a new candidate is registered in the list, the system sends a notification to the administrator stating, "A new automation rule candidate has been discovered. Please review and apply it." Additionally, an interface is provided that allows the administrator to check the candidate list on the administrator page and easily add the content as a new rule to the 'keyword dictionary' of the text analysis module (200) or the 'rule table' of the rule-based processing module (300) with a single click.

[0212] As a specific scenario, the processing procedure for a situation in which orders containing a new request for "10% improvement in heat dissipation performance" are repeatedly received is performed step by step. In the first step, on January 1, the text analysis module (200) fails to interpret the phrase and records it in the "unclassified pattern log." In the second step, the person in charge checks the request, internally adds a specific part, and processes it as "successfully completed," and the system records this processing result along with the log. In the third step, the same request is received four more times during January, and the person in charge consistently adds the same part each time and processes it as "successfully completed." In the fourth step, the moment the fifth valid processing history is recorded on January 30, the system determines that the "repetitiveness" criterion of five or more times within one month has been satisfied. In the final fifth step, the system automatically registers the input pattern and processing method as a pair in the "rule update candidate list" and sends a notification to the system administrator requesting a review of the content.

[0213] The identification method of the present invention is configured to maximize the reliability of the logic by adopting a human-system collaboration approach. The system focuses on the role of 'discovering' and 'proposing' learning targets, while the authority to ultimately 'apply' new rules to the system rests strictly with the human administrator. This acts as a key safeguard to fundamentally prevent the system from automatically learning incomplete or incorrect patterns that could adversely affect the entire system, and the administrator controls the process by permanently designating specific patterns from the candidate list as 'excluded from learning,' thereby explicitly managing exceptional cases where automation is inappropriate.

[0214] When comparing the present invention with the prior art static rule-based system, in the same situation where requests for improved heat dissipation performance occur repeatedly, the prior art system fails to process the request every time and classifies it as an exception, requiring the person in charge to repeat the same manual work every time, and the system's automation rate remains unchanged over time. In contrast, the present invention requires the person in charge to handle the request manually for the first few times, but the system learns this pattern and proposes the addition of a new rule to the administrator upon the fifth occurrence. Once the administrator approves the rule, the system switches to processing the request 100% automatically. As a result, the present invention allows the business knowledge of field personnel to naturally be converted into system rules, enabling a rapid response to changes and achieving the effect of the system evolving on its own without separate development costs.

[0215] The judgment rule is set so that if the system detects a pattern that cannot be processed in step 1, it is recorded in the 'unclassified pattern log' and proceeds to step 2; if the person in charge processes it as 'successfully completed' and the consistency is 80% or higher, it is registered as a 'potential rule candidate' and proceeds to step 3; and if the 'potential rule candidate' occurs more than 5 times within a threshold of 1 month in step 3, it is finally selected as a 'rule update candidate'.

[0216] The judgment rules of the present invention are configured to be technically grounded in the '80 / 20 rule' for efficient system maintenance. That is, since attempts to automate all exception cases are inefficient, the 'repetitiveness' criterion of the present invention identifies only core patterns that occur repeatedly and increase the workload of personnel, designating them as targets for automation. This serves as an objective and inevitable strategy to maximize the overall automation rate of the system by utilizing limited maintenance resources most effectively.

[0217] The adoption of each component has specific technical effects. By adopting new pattern identification criteria, the system recognizes its own knowledge limitations and establishes a starting point for continuous improvement. By adopting manual processing validity criteria, it executes commands that ensure the reliability of the learning process and prevent system contamination caused by incorrect information by targeting only knowledge verified by human experts. By adopting pattern repeatability criteria, it acts as a filter to evaluate the economic utility of automation, controlling the system to achieve maximum efficiency improvement with minimal effort. Through these sequential gate-based rules, the system evolves beyond a simple automated processing tool into a dynamic platform that learns and grows on its own, achieving technical effects that reduce long-term operating costs and maximize the system's adaptability and flexibility to changing market environments.

[0218] The terms and threshold criteria used in this invention are clearly defined. 'New order patterns' refer to all text phrases or request combinations that are not processed by existing keyword dictionaries, rule tables, regular expressions, etc. 'Consistency of manual processing' measures text similarity between multiple manual processing result data and is configured to be quantitatively evaluated through sentence embedding and cosine similarity calculation methods. Finally, thresholds are set by reflecting an empirical reference point determined through consultation between the operations manager and field personnel at the initial stage of system implementation—a threshold deemed worthy of being established as a rule after a certain level of repetition—and provide a function to adjust them at any time through the system administrator tool.

[0219] The threshold value of the above 'repetitiveness' criterion (5 times within 1 month) is set through an automated break-even point (BEP) analysis of 10 representative new patterns that occurred over the past year. The purpose of the simulation is to find the minimum number of iterations that is economically meaningful, such that the cost wasted due to repetitive manual processing becomes greater than the cost incurred in generating rules.

[0220] [Table 4-1]

[0221]

[0222] As a result of the simulation above, the break-even point for various new patterns was found to be approximately 6 times on average. In particular, patterns occurring 5 or more times per month on average (numbers 1, 3, 5, 7, 9, and 10) mostly approached or exceeded the break-even point, indicating that the economic utility of automation is high. On the other hand, for patterns occurring 4 times or less per month, manual processing may be more cost-effective. Therefore, based on this break-even point analysis, the present invention is characterized by controlling system maintenance resources so that they are allocated to the most efficient location by setting 5 times, which ensures economic feasibility in most cases, as the monthly repeatability threshold.

[0223] To verify the effectiveness of the set threshold (5 times / month), a simulation was performed on 50 new patterns that occurred over the past 6 months. This simulation evaluates how well the system selects patterns with high automation value (efficiency) and whether it hinders managers with unnecessary rule suggestions (accuracy).

[0224] [Table 4-2]

[0225]

[0226] As a result of the verification simulation above, the system accurately identified 13 out of 15 patterns requiring automation, demonstrating a high valid proposal selection rate of 86.7%. This implies that the system does not miss key repetitive tasks that can significantly improve work efficiency. Furthermore, 13 out of the 15 patterns proposed by the system were found to actually require automation, recording a proposal accuracy of 86.7%. This proves that the system effectively filters out issues, preventing managers from wasting time reviewing unnecessary and worthless proposals. These metrics are calculated using the following conceptual formula.

[0227] Valid Proposal Selection Rate (%) = [Number of actual automation-required patterns proposed by the system] / [Total number of actual automation-required patterns] × 100

[0228] = 13 / 15 X 100 = 86.7%

[0229] Proposal Accuracy (%) = [Number of actual patterns requiring automation suggested by the system] / [Total number of patterns suggested by the system] × 100

[0230] = 13 / (2 + 13) Х 100 = 86.7%

[0232] As an embodiment of the present invention, the entire processing process is configured to be explained in detail when an order is received for lighting fixtures to be installed at the new project site, the 'Ocean Palace Hotel external coastal promenade'.

[0233] The first step is the data reception and original text analysis step, during which an email is received by the system with the following content: "We are ordering 30 units of the model name OL-3000 stand type from an external partner. The site is the coastal promenade outside the Ocean Palace Hotel. Please ensure proper waterproofing, and please position the terminals appropriately so they are not visible from below." The data reception unit (100) receives the entire text of the email body as raw data and immediately transmits it to the text analysis module (200). The text analysis module (200) performs Process 1 (order ambiguity grade determination process) on the received text, identifying three expressions such as "proper waterproofing," "not visible from below," and "appropriately" as "non-standard expressions" and assigning a risk score of 3 points, and confirming that "color temperature" and "foundation anchor set" are missing as "missing essential information" and assigning a risk score of 8 points. No obvious 'conflict of meaning' is found, so 0 points are assigned. The total risk score is summed to 11 points, and since this exceeds the threshold of 10 points, a command is executed to determine the 'order ambiguity grade' as 'ambiguous'.

[0234] The second step is the potential accessory identification step, in which the rule-based processing module (300) receives the text and the judgment result and performs process 2 (automatic recommendation process for potential accessories). Based on the 'technical relevance' determination result, it is confirmed that the 'foundation anchor set' is 'essential', and a necessity score of 100 points is assigned to determine it as 'essential' grade. Based on the 'site environment requirements' determination result, external and coastal keywords are extracted and the relevant table is queried to assign necessity scores of 50 points to 'SUS316 grade bolt', 20 points to 'salt-resistant coating', and 20 points to 'surge protector (SPD)', and all three items are determined as 'recommended' grade. Based on the 'statistical recommendation' determination result, it is processed as having no significant pattern.

[0235] The third step is the automatic processing reliability determination step, where the system combines the results of the previous two steps to execute Process 3 (automatic processing reliability determination logic). It receives the 'Ambiguous' grade, which is the result of Process 1, calculates the 'Intervention Score' based on the result of Process 2 to yield a total of 35 points, and evaluates the 'System Intervention Level' as 'High' based on this. Referring to the 'Judgment Rule Matrix,' since the 'Source Text Clarity Grade' is 'Ambiguous' and the 'System Intervention Level' is 'High,' it executes a command to finally determine the 'Automatic Processing Reliability' as 'Low.'

[0236] The fourth step is the user verification and final confirmation step, in which control of the system is transferred to the user verification module (600) according to the determined 'automatic processing reliability' grade. Since the reliability grade is 'low', the system stops all automatic processing procedures and sends a 'required verification and approval' request to the ordering person's interface. The request clearly specifies items requiring verification, such as 'terminal location designation required' and 'color temperature value missing', as well as 'foundation anchor set' automatically included by the system, and recommended items such as 'SUS316 grade bolt', 'salt-resistant coating', and 'surge protector', along with the reasons for them. The ordering person verifies the above information, clearly inputs the terminal location as 'bottom' and the color temperature as '4000K' through the system interface, adopts all three recommended items, and responds by clicking the final 'approval' button.

[0237] The fifth step is the data recording and rule learning preparation step, in which the data finally confirmed by the person in charge is recorded in the data storage unit (700). First, all modified and confirmed specifications are stored as official order data. Then, according to process 4 (method for identifying pattern targets for rule learning), the system groups the processing results of the original text external coastal promenade and the finally added accessories into a pair and records them as 'potential rule candidates', and executes a command to increase the occurrence count by 1.

[0238] The final sixth step is the final document generation step, and once the user's final approval is complete, control is transferred to the document generation module (500). Based on the confirmed final order data, the module automatically generates a production order with all specifications clearly stated and transmits it to the production management system. Through the above process, the initial order request, which was ambiguous and incomplete, passes through each component and process of the present invention, thereby eliminating all potential risks and completing the process of automatically converting it into a clear production order complete with expert-level specifications.

[0239] The technology according to the present invention, as described above, may be implemented in the form of program instructions that can be executed through various computer components or implemented as an application, and may be recorded on a computer-readable recording medium. The computer-readable recording medium may include program instructions, data files, data structures, etc., either individually or in combination.

[0240] The program instructions recorded on the above-mentioned computer-readable recording medium are those specifically designed and configured for the present invention, but may also be those known and available to those skilled in the art of computer software.

[0241] Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical recording media such as CD-ROMs and DVDs; magneto-optical media such as floptical disks; and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, and flash memory.

[0242] Examples of program instructions include machine code, such as that generated by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc. The hardware device may be configured to operate as one or more software modules to perform processing according to the present invention, and vice versa.

[0243] Although the invention has been described above with reference to embodiments, those skilled in the art will understand that various modifications and changes can be made to the invention without departing from the spirit and scope of the invention as set forth in the following claims.

Claims

Claim 1 A drawing-based automatic specification extraction and verification ordering system that automatically extracts dimensions, structural information, and specification information from drawings and specification images, performs verification and correction by comparing the extracted information with a standard catalog, and automatically generates wiring connection guides and order forms, wherein the drawing-based automatic specification extraction and verification ordering system comprises: a data analysis unit that recognizes dimension lines, arrows, labels, and symbols from a drawing image to extract structural information including dimensions, height, hole pattern, and structural information of a lighting fixture, and extracts specification information including material information, luminous flux, color temperature, and power specifications from a table or text area within a specification image; a normalization unit configured to compare the structural information and specification information extracted by the data analysis unit with standard catalog information, determine the suitability of the extracted values ​​by considering an allowable error range set by manufacturer or product family, and automatically present alternatives having the same luminous flux group, the same color temperature group, or equivalent specifications if the values ​​exceed the allowable error; and a normalization unit configured to recognize wiring diagrams or wiring symbols within the drawing image to interpret wiring connection information including terminal location, polarity, and grounding status, and automatically generate wiring guide information based on the interpreted wiring connection information, and a request for terminal location change A drawing-based automatic specification extraction and verification ordering system comprising: a scenario management unit configured to provide terminal location change options for frequent models; and an automatic order generation unit that determines whether the inclusion of a mounting bracket or accessory is necessary based on the structural information, specification information, and wiring connection information, recommends basic inclusion, and generates an order template with site name, quantity, and delivery date information automatically filled in. Claim 2 delete Claim 3 A drawing-based automatic specification extraction and verification ordering system, wherein the normalization unit calculates a normalized value for each of a plurality of specification items, including luminous flux, color temperature, power consumption, power specifications, external dimensions, and weight, among the specification information extracted by the data analysis unit, by referring to a predefined or learned tolerance range and weighting for each manufacturer or product family. Claim 4 In paragraph 3, the normalized value for each item is calculated as a dimensionless deviation value obtained by dividing the difference between the extracted specification value and the corresponding specification value in the standard catalog by the allowable error range; the specification item is determined to be suitable if the dimensionless deviation value is below a reference threshold, and is determined to be unsuitable if it exceeds the reference threshold; the normalization unit is configured to calculate a suitability score for each specification item by applying the weight to the normalized value for each item, and to calculate a comprehensive normalized similarity score for each candidate product in the standard catalog by summing or weighting the suitability scores for each specification item; based on the comprehensive normalized similarity score, multiple candidate products in the standard catalog are rearranged in order of similarity, and candidate products with a comprehensive normalized similarity score above a reference value are selected as a suitable candidate group; the normalization unit is configured to automatically identify as a replacement candidate a product in which the luminous flux or color temperature belongs to the same group and the power consumption or power specification is included within the allowable error range by comparing the extracted specification information among the suitable candidate group, and for the replacement candidate, a replacement including the difference item compared to the original specification, the degree of difference, and whether the allowable error is satisfied A drawing-based automatic specification extraction and verification ordering system configured to generate reason information together, wherein the normalization unit is configured to classify a candidate as a restricted suitable candidate and provide it to the user for selection, even if the comprehensive normalization similarity score is below a reference value, if the normalization value of a pre-set core specification item among the item-specific normalization values ​​is included within the reference range. Claim 5 A drawing-based automatic specification extraction and verification ordering system, wherein the data analysis unit identifies dimension line candidates based on length, direction, and repeating pattern among straight line segments from an uploaded drawing image, detects arrow-shaped objects placed near both ends of the dimension line candidates, determines whether the direction of the centerline direction vector of the dimension line and the tip direction vector of the arrow-shaped objects match for each dimension line candidate, determines the dimension line indicating the actual dimension, detects a dimension text label placed in correspondence with the determined dimension line, establishes a correspondence relationship based on the normal direction separation distance and parallelism between the dimension text label and the dimension line, and calculates the scale factor of the drawing image by correlating a reference length interval defined by the pixel distance between the arrows at both ends of the dimension line with the actual dimension value listed in the dimension text label. Claim 6 In claim 5, the data analysis unit, when calculating the scale factor, derives a plurality of dimension line candidates corresponding to the dimension text label by utilizing a relative positional relationship including the pixel distance between the arrows at both ends of the dimension line, the distance at which the dimension text label is separated in the normal direction of the dimension line, and whether there is an intersection or overlap between the dimension line and an adjacent line segment; for each of the plurality of dimension line candidates, calculates a scale factor candidate based on the pixel length of the dimension line and the dimension value of the dimension text label corresponding to it; calculates an average value and a variance value for the calculated plurality of scale factor candidates; removes scale factor candidates whose variance value exceeds a preset reference variance value as outliers; determines the average value or median value of the remaining scale factor candidates as a representative scale factor; calculates structural information by converting the pixel length measured in the image coordinate system into an actual measurement unit for at least one structural element among the hole center point, hole pitch, outer contour length, height dimension, or mounting surface reference separation distance within the drawing image based on the representative scale factor; and if the structural information exceeds the allowable error range when compared with the structural information listed in the standard catalog, A drawing-based automatic specification extraction and verification ordering system, wherein the data analysis unit compares multiple dimension line direction vectors within a drawing image and determines that rotational distortion exists if the dimension line direction vectors are not parallel to each other, applies a rotational correction transformation to the drawing image based on the average direction of the dimension line direction vectors, and if the difference between the scale factor candidates calculated from the horizontal dimension line and the vertical dimension line exceeds a reference difference value even after the rotational correction, recalculates independent scale factors for the horizontal and vertical directions respectively, and applies the recalculated scale factors to re-correct and output the structural information.

Images