A PCB pricing and customer service integrated system and method based on artificial intelligence

Abstract

The application relates to the field of artificial intelligence and data processing technology, and discloses a PCB pricing and customer service integrated system and method based on artificial intelligence, aiming to solve the problems of low efficiency and dependence on manual work in PCB industry pricing, and the problem of the existing intelligent customer service and pricing system being fragmented and unable to handle complex process requirements. The method comprises the following steps: a user describes requirements in natural language through a dialogue interface; the system processes user input through a hybrid scheduling mechanism and extracts key process parameters to convert into structured data; when the parameters meet the conditions or the user issues an instruction, a pricing engine is called to calculate and feed back the results according to a preset model; and interaction data is recorded and associated with a user account. The system comprises the following modules: a pricing rule database, a core pricing engine, an AI interface gateway, an AI supplementary instruction library, natural language processing and scheduling, a user account and data record, and a unified user interface. Through the above technical scheme, the application can realize second-level automatic pricing of complex PCBs, improve efficiency and accuracy, and form a complete business closed loop of requirement acquisition, real-time pricing and data sedimentation.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence and data processing technology, specifically relating to an integrated system and method for PCB pricing and customer service based on artificial intelligence. Background Technology

[0002] With the continuous development of the electronics manufacturing industry, printed circuit boards (PCBs), as core components of electronic products, are highly customized and have complex process parameters, making the quotation process a key link affecting customer experience and enterprise response efficiency. Currently, PCB pricing generally relies on manual intervention, requiring experienced engineers to perform item-by-item calculations based on multi-dimensional parameters such as size, number of layers, materials, and special processes provided by the customer, combined with internal cost models. This process is not only time-consuming and error-prone, but also difficult to standardize. Although some manufacturers have deployed online quotation tools, these are usually only applicable to standard board types with simple structures and standard processes. Once complex requirements such as blind and buried vias, thick copper, high-frequency materials, or composite surface treatments are involved, the system cannot accurately analyze them and still needs to be handed over to manual processing, which seriously restricts the level of service automation.

[0003] Among these, AI-based customer service and pricing integration technology has become an important direction for improving industry efficiency. An ideal technical solution should be able to understand the process requirements described by customers in natural language, automatically extract key parameters, and generate accurate quotes in real time by calling professional pricing models, while supporting contextually coherent multi-turn interactions. However, existing customer service systems mostly adopt a general chatbot architecture, lacking the ability to recognize PCB industry terminology and process logic. They cannot accurately identify professional expressions such as "immersion gold," "impedance control," and "resin via plugging," nor can they dynamically accumulate parameter information during dialogue to drive the pricing engine. This results in superficial interactions that are difficult to support a real business closed loop.

[0004] Existing technologies exhibit a clear disconnect between intelligent customer service and professional pricing systems: on the one hand, while traditional pricing systems possess rigorous rules, their reliance on structured forms for input results in a rigid user experience; on the other hand, while large language models offer advantages in natural language understanding, they lack in-depth modeling of PCB cost composition, making direct application to pricing prone to inaccuracies and failing to guarantee consistency and compliance in responses to frequently asked standard questions. Furthermore, the demand descriptions, parameter extraction records, and historical pricing data generated during user interactions are not effectively collected and correlated, making it difficult for companies to extract customer behavior characteristics and market trends. Therefore, achieving a deep integration of natural language interaction, dynamic parameter extraction, rule-based pricing, and data asset management, while ensuring pricing accuracy and professionalism, has become a pressing technical challenge for intelligent services in the PCB industry. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide an integrated system and method for PCB pricing and customer service based on artificial intelligence. This system can effectively solve the technical problems in the background art, such as PCB pricing relying on manual labor, low efficiency, the separation of existing intelligent customer service and pricing systems, inability to handle complex process requirements, and the lack of effective accumulation and utilization of interactive data.

[0006] To achieve the above objectives, the present invention provides the following technical solution: On one hand, an AI-based PCB pricing and customer service integration system, comprising the following components: a pricing rule database module, used to store PCB price calculation models constructed based on multi-dimensional process factors, wherein the multi-dimensional process factors include size area, quantity, material type, number of layers, copper thickness, surface treatment process type, number of holes, number of test points, minimum hole diameter, minimum line width and spacing, board utilization rate, expedited marking, and other special processes; the pricing logic of the price calculation model is: the sum of the basic area cost and the costs of each process factor, wherein the basic area cost is priced in segments according to a preset area range, and the costs of each process factor consist of a fixed surcharge per item and a surcharge per square meter; a core pricing engine module, used to receive structured process parameter data, call the price calculation model in the pricing rule database module to perform calculations, and output the price results; and an AI interface gateway module, providing a configurable application programming interface for accessing at least one external AI large language model, through which the system administrator configures the currently active AI model instance; The AI ​​supplementary instruction library module stores user-defined AI supplementary instruction pairs. Each instruction includes a question and a preset response. When the user's natural language question matches the question, the system prioritizes returning the corresponding preset response. The natural language processing and scheduling module receives and processes the user's natural language input. This module executes the following logic: First, it performs a similarity match between the user's question and the question in the AI ​​supplementary instruction library module. If the match is successful, it directly returns the corresponding preset response. If the match fails, it sends the user's question and the current conversation context to the configured AI model through the AI ​​interface gateway module and returns the AI ​​model's response to the user. This module also integrates an information extraction submodule, which identifies and extracts key process parameters affecting pricing from continuous dialogues and converts them into structured data that can be processed by the core pricing engine module. The user account and data recording module provides user authentication and login functions and records the query behavior data of logged-in users. The query behavior data includes original dialogue records, system response content, triggered pricing requests and their results, query timestamps, and query count statistics. Users can view the historical data bound to their accounts through the backend. The unified user interface module provides a conversational customer service interface and a backend data query interface. The conversational customer service interface supports natural language interaction in the form of a chatbot, and the backend data query interface allows logged-in users to retrieve and analyze their historical interactions and pricing records.

[0007] Preferably, the price calculation model stored in the pricing rule database module has its pricing logic precisely expressed by the following formula: in, Represents the total price. Represents the area of ​​the PCB. Representation based on area The basic area cost function is determined by the interval in which it is located. This represents the total number of technological factors that influence price. Representing the Each process factor incurs a fixed additional charge. Representing the Additional charges per square meter for process factors. The aforementioned basic area cost function. Defined as a piecewise function, for example, when the area When it is less than 0.05㎡, ;when When the area is in the range of [0.05, 0.1)㎡, ;when When it is greater than or equal to 0.1㎡, ,in , , The preset base price per square meter and .

[0008] Furthermore, the similarity matching in the AI ​​supplementary instruction library module is implemented using a cosine similarity algorithm based on word vectors or semantic similarity calculation based on a pre-trained language model. A similarity threshold is set. (For example When a user's question is similar to any question in the database, the similarity score is calculated. satisfy When the match is successful, the question and response content can be added, deleted, modified, and queried through the system management backend to ensure that highly frequent, standard, or compliance-critical questions receive absolutely consistent and accurate responses.

[0009] Furthermore, the information extraction submodule within the natural language processing and scheduling module employs a combination of named entity recognition and relation extraction techniques. This submodule incorporates a PCB-specific entity dictionary, including technical terms such as "immersion gold," "impedance control," "blind vias," "resin plugs," "FR4," and "high-frequency materials," along with their synonym mappings. During the dialogue, this submodule scans user statements in real-time, identifies and labels entities corresponding to preset process parameter categories, and maintains a dynamically updated structured data sheet through contextual association across multiple rounds of dialogue. This data sheet stores parameters in key-value pair format.

[0010] On another front, an AI-based PCB pricing and customer service integration method is described, comprising the following steps: Step S110: The user initiates an interactive session through the conversational customer service interface of the unified user interface module. The user can choose to interact as logged in or anonymously. Step S120: The user describes their PCB requirements in natural language, including one or more parameters such as size, number of layers, material, and process requirements. Step S130: The natural language processing and scheduling module receives user input and executes scheduling logic: it performs similarity matching between the user input and the AI ​​supplementary instruction library. If the match is successful, it returns a preset reply and ends the current processing branch. If the match fails, it sends the user input and the current session context to an external AI model through the AI ​​interface gateway to obtain an intelligent reply, while activating the information extraction submodule. Step S140: The information extraction submodule continues to run during the dialogue processing in step S130, extracting key process parameters from the user's historical and current statements and filling them into a structured data sheet. Step S150: When the parameters in the structured data sheet meet the preset pricing trigger conditions, or when the user explicitly issues a pricing instruction, the core pricing engine module is invoked. This module reads the structured parameters in the data sheet and performs calculations based on the model in the pricing rule database. Step S160: The system replies to the user in natural language through the conversational customer service interface with the pricing result calculated in step S150, along with necessary explanations of the cost composition. Step S170: Throughout the entire interaction process from steps S110 to S160, the system temporarily records the original dialogue records, extracted structured parameters, pricing requests, and results. If the user is logged in, the complete record is bound to the user account and persistently stored in the user account and data recording module. Step S180: Logged-in users can retrieve, view, and analyze the historical records bound to their accounts through the background data query interface of the unified user interface module, based on time range, project keywords, or process parameters.

[0011] Preferably, the parameter extraction in step S140 employs an attention-based sequence labeling model. This model is trained on a large dataset of labeled PCB requirement dialogues and can accurately identify parameter values ​​and their categories in unstructured descriptions. For example, for the input statement "board thickness 1.6 mm, surface treatment is tin plating", the model will output a labeling sequence, identifying "1.6" as belonging to the "board thickness" category and "tin plating" as belonging to the "surface treatment process" category.

[0012] Furthermore, the pricing triggering conditions in step S150 are either predefined by the system or dynamically determined. The predefined condition is that pricing is automatically triggered when the structured data sheet contains the five core parameters: "size / area," "number of layers," "quantity," "material," and "surface treatment process." The dynamic determination condition is that by analyzing the user's intent, pricing is triggered even if the parameters are incomplete when key intent words such as "quote," "how much," or "calculate price" are identified. The AI ​​model then proactively queries for missing parameters in the response.

[0013] Compared with the prior art, the present invention has the following beneficial effects: 1. By digitizing professional PCB pricing rules into precise mathematical models and integrating them with the core pricing engine, we have achieved second-level automated pricing for complex process boards, completely eliminating the delays and errors of manual calculation and significantly improving pricing efficiency and accuracy.

[0014] 2. By integrating an AI-supplemented instruction library with an external large language model through a hybrid scheduling mechanism, the system ensures absolute consistency and accuracy in responses to high-frequency, standard, and compliant questions, while fully utilizing the natural language understanding and generalization capabilities of the large model to handle open-ended questions, thus achieving a natural interactive experience that is both intelligent and controllable.

[0015] 3. Through the natural language processing and scheduling module and its integrated information extraction sub-module, it innovatively completes demand understanding, parameter extraction and structured transformation in real time in the dialogue flow, and seamlessly connects with the pricing engine and data recording module to form a complete business closed loop of "demand acquisition - real-time pricing - data accumulation", solving the problem of system fragmentation.

[0016] 4. All core components of the system are designed as configurable modules. Pricing rules, AI model access, and supplementary instruction libraries can all be dynamically maintained and updated through the management backend, enabling the system to flexibly adapt to changes in raw material cost fluctuations, production process innovations, and common customer inquiries, thus possessing long-term applicability.

[0017] By using user accounts and data recording modules to structure and link interactive data throughout the entire process, enterprises accumulate high-value customer demand data assets. This data can be used for in-depth analysis of market trends, optimization of production processes, and implementation of precise customer service and marketing strategies, thereby enhancing the value of the data. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall technical solution architecture of the PCB pricing and customer service integration system based on artificial intelligence proposed in this invention; Figure 2This is a schematic diagram of the core principle framework of the hybrid scheduling mechanism that integrates AI supplementary instruction library and external large language model in this invention; Figure 3 This is a logical flow framework diagram of the complete business loop of "demand acquisition - real-time pricing - data accumulation" in this invention; Detailed Implementation

[0019] 1. To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments. Example 1

[0020] In the electronics manufacturing services industry, PCB suppliers' sales and technical support departments handle a large number of inquiries and consultations from various customers daily. Customer needs vary widely, ranging from simple "how much does a four-layer board cost?" to customized requests involving dozens of complex process parameters. Traditionally, customer service personnel must first understand the customer's needs, then manually enter the required parameters into separate pricing software or spreadsheets, calculate the price, and provide feedback to the customer. This process is not only time-consuming and labor-intensive but also prone to inaccurate quotations due to misunderstandings or incorrect parameter entry. Furthermore, a large amount of customer interaction data is scattered across chat logs and emails, failing to form a structured data asset for analysis. The AI-based PCB pricing and customer service integration system provided by this invention is designed to address this specific business scenario, aiming to achieve full automation and intelligence from natural language dialogue to accurate quotations and data accumulation.

[0021] See Figure 1 The overall architecture of this system includes a pricing rule database module, a core pricing engine module, an AI interface gateway module, an AI supplementary instruction library module, a natural language processing and scheduling module, a user account and data recording module, and a unified user interface module. These modules work together to form a complete closed loop from demand input to result output.

[0022] The pricing rules database module is the cornerstone of the entire system's pricing structure. It stores a highly structured PCB price calculation model built upon multi-dimensional process factors. This model decomposes the cost composition of PCB manufacturing into quantifiable expense items. These multi-dimensional process factors include, but are not limited to: size / area, quantity, material type, number of layers, copper thickness, surface treatment process type, number of holes, number of test points, minimum hole diameter, minimum line width / spacing, board utilization rate, rush processing markings, and other special processes. The pricing logic of the model is defined as follows: the total price equals the sum of the base area cost and the costs of each process factor. The base area cost is priced in segments based on a preset area range to reflect the marginal cost differences under different production scales. The costs of each process factor consist of two parts: a fixed surcharge per item and a surcharge per square meter. The fixed cost covers the fixed costs of process switching or special treatments, while the cost per square meter is linearly related to the PCB area, reflecting the area correlation of material or processing costs. This pricing logic is precisely expressed through the following mathematical formula: in, Represents the total price. Represents the area of ​​the PCB. Representation based on area The basic area cost function is determined by the interval in which it is located. This represents the total number of technological factors that influence price. Representing the Each process factor incurs a fixed additional charge. Representing the Additional charges per square meter for various process factors. Basic area cost function. It is explicitly defined as a piecewise function, for example, when the area When it is less than 0.05㎡, ;when When the area is in the range of [0.05, 0.1)㎡, ;when When it is greater than or equal to 0.1㎡, .in , , The preset base price per square meter, and follows The pricing rules database module is designed to reflect the scale effect of lower basic costs per unit area for larger areas. It is maintained through a management backend, allowing administrators to dynamically adjust various cost parameters, add or delete process factors, or modify segment intervals to ensure the pricing model remains synchronized with real-time market costs and production capacity.

[0023] The core pricing engine module is the "brain" that performs the calculations. It receives structured process parameter data from the natural language processing and scheduling module. This data is organized in key-value pairs, such as {"size / area": ​​0.08, "number of layers": 4, "material": "FR4", "surface treatment": "tin plating"}. The core pricing engine module first parses these parameters, identifying the valid fields related to pricing. Then, it calls the pricing rules database module to determine the applicable base area cost function based on the "size / area" parameter. Then, iterate through all provided process parameters and retrieve the corresponding fixed surcharge from the database for each parameter. And additional charges per square meter For each item, calculate its cost contribution. Finally, add all the cost items to the base cost to get the final total price. This module has built-in strict parameter validation logic. For missing necessary parameters or parameter values ​​that exceed the process capability range, it will generate clear error codes and prompts, and return them through the unified user interface module, instead of performing incorrect calculations.

[0024] The interface gateway module bridges the system to external AI capabilities. It's a configurable application programming interface (API) layer designed to connect to at least one external large-scale AI language model service. System administrators can input connection parameters such as the target AI model's API endpoint address, authentication key, request timeout, and maximum token count through a dedicated configuration interface. This module encapsulates the communication details with APIs from different vendors, converting the internally unified request format to the specific format required by the target model and handling response parsing. This design allows the system to flexibly switch between or use multiple AI models simultaneously. For example, general question-and-answer requests with high response speed requirements can be routed to a lightweight model, while complex technical inquiries requiring deep reasoning can be routed to a more powerful model. The AI ​​interface gateway module also monitors the availability and response latency of AI services and executes failover or degradation strategies in case of service anomalies.

[0025] The supplementary instruction library module stores and manages user-defined AI supplementary instruction pairs. Each instruction pair contains two core fields: the question content and the preset response content. The question content defines the user's question pattern that triggers the instruction; it can be an exact string or an expression containing wildcards or semantic templates. The preset response content is a pre-approved standard answer that the system will directly return after a successful match. Application scenarios for this module include: answering company-specific policy questions, providing official explanations of standard processes, responding to frequently asked simple inquiries, and ensuring absolute consistency in answers to questions with extremely high compliance requirements. System administrators can add, delete, modify, and query instructions through the backend management interface. When performing matching, the system uses a cosine similarity algorithm based on word vectors or a semantic similarity calculation based on a pre-trained language model. The system sets a similarity threshold. ,For example The similarity score between the natural language question entered by the user and any question in the database. satisfy Upon successful matching, the system will interrupt subsequent AI model calls and directly return the corresponding preset response. This ensures the accuracy and controllability of key information.

[0026] The Natural Language Processing and Scheduling module is the core hub for processing user natural language input; its internal logic can be found in [link to internal logic]. Figure 2 This module first receives raw user input text from the unified user interface module. The first step is to execute the scheduling logic: performing real-time similarity matching calculations between the user's question and all queries in the AI ​​supplementary instruction library module. If a match is successful, the corresponding preset response is immediately retrieved from the library, packaged into a system response, and returned directly to the user, ending the current processing branch. If a match fails, indicating that the user's question is open-ended or non-standard, the scheduling logic initiates a second path: sending the user's question and the context information of the current session through the AI ​​interface gateway module to the currently configured and activated external AI model instance. The AI ​​model, based on its powerful natural language understanding and generation capabilities, generates an intelligent response. This response is received by the natural language processing and scheduling module and returned to the user. Meanwhile, regardless of whether a match is successful or not, as long as the dialogue continues, the information extraction submodule integrated within this module will remain active and running.

[0027] The information extraction submodule is specifically responsible for identifying and extracting key process parameters affecting PCB pricing from continuous, unstructured dialogue text, and converting them into structured data that can be processed by the core pricing engine module. This submodule adopts a technical approach combining named entity recognition and relation extraction. Its underlying layer incorporates a meticulously constructed PCB-specific entity dictionary, which not only includes core technical terms such as "immersion gold," "impedance control," "blind vias," "resin-filled vias," "FR4," "high-frequency materials," and "Rogers board," but also establishes a synonym and abbreviation mapping table; for example, "ENIG" maps to "chemical immersion gold," and "OSP" maps to "organic solder mask." During the dialogue, the information extraction submodule scans each user statement in real time, using an attention-based sequence labeling model to analyze the text word by word. This model has been thoroughly trained on a large corpus of manually labeled PCB requirement dialogues, enabling it to accurately identify parameter values ​​and their corresponding process categories in unstructured descriptions. For example, given the input statement "I need a four-layer board with a thickness of 1.6 mm, a length and width of 10 cm by 5 cm, and a surface treatment of tin plating," the model will output a label sequence, identifying that "1.6" belongs to the "board thickness" category, "10 cm" and "5 cm" both point to the "size" category, "tin plating" belongs to the "surface treatment process" category, and "four layers" belongs to the "number of layers" category. The identified entities and their categories are extracted. This submodule maintains a dynamically updated structured data sheet, which stores parameters in key-value pairs. In multi-turn dialogues, the submodule can handle parameter clarification, modification, and supplementation through contextual analysis. For example, if a user first says "make a four-layer board" and then adds "no, change it to six layers," the submodule will update the value of the "number of layers" field in the data sheet from "4" to "6." The fields in the data sheet correspond to the list of process factors in the pricing rules database module.

[0028] The user account and data logging module provides standard user authentication and login functionality. Users can register an account and log in to the system. The core function of this module is to comprehensively record all interaction data between each logged-in user and the system, forming a valuable "data repository." Recorded data items include: original user dialogue records, the content of every system response, all pricing requests triggered during the session and their detailed calculation results, the precise timestamp of each interaction, and the user's cumulative query count statistics. This data is not stored in isolation but is strongly linked through a unique user identifier. After a user logs in, all subsequent session data is automatically bound to their account and persistently stored in the system's database. This allows businesses to trace the complete evolution of a user's needs and their inquiry history on a user-by-user basis.

[0029] The unified user interface module serves as the central entry point for system-user interaction, offering two main interface formats. The first is a conversational customer service interface, embedded as a chatbot within the company's website, mobile application, or instant messaging tool. Users can interact with this interface as if speaking to a live customer service representative, using natural language to describe their needs, ask questions, and request quotes. All intelligent responses, quote results, and parameter confirmation inquiries are presented to the user through this interface in natural language, providing a smooth interactive experience. The second is a backend data query interface, a web-based management interface for logged-in users. After logging in, users can use this interface to retrieve, view, and analyze all historical interactions and pricing records linked to their account. This interface supports filtering and querying by time range, project keywords, or specific process parameters. For example, a user can query "all inquiry records involving 'impedance control' processes within the past three months," and the system will list relevant historical conversations and quote details, supporting export as reports for easy project review or cost analysis.

[0030] Based on the above system, this embodiment details the execution flow of an AI-based PCB pricing and customer service integration method. (See attached document for details.) Figure 3 The specific steps are as follows: In step S110, the user initiates an interactive session through the conversational customer service interface of the unified user interface module. The user interface is presented as a simple chat window. The user can choose to start the conversation directly as an anonymous guest, or click the login button, enter a username and password to complete authentication, and interact as a logged-in user. The system will record the initial session identifier and bind it to the user's identity.

[0031] In step S120, the user describes their PCB requirements in natural language through the input box in the chat window. The requirement description can be fragmented and conversational, such as: "Hello, I want to make a circuit board," "The board is approximately 10 cm long and 5 cm wide," "Ordinary FR4 material is fine," "It needs to be four layers, with immersion gold surface treatment," "Also, there are several locations that require impedance control, 90 ohms differential." The user may state multiple parameters in a single sentence or gradually supplement information through multiple rounds of dialogue.

[0032] Step S130: The natural language processing and scheduling module receives the user's input text in real time. The module immediately initiates the scheduling logic: it performs a fast similarity matching calculation between the user's input sentence and the question content of all entries in the AI ​​supplementary instruction library. For example, if the user input is "What is your company's typical delivery time?", and there happens to be a question in the instruction library with the same question content, the similarity score is calculated. Above the threshold If the user inputs "My board has blind vias, will it significantly affect the price?", then the matching is successful. The system directly returns the preset response content corresponding to the instruction, such as "Dear customer, our standard delivery time is 7-10 working days. Rush orders can be negotiated to 3-5 working days, depending on the complexity of the board's manufacturing process." and ends the current round of processing. If the user inputs "My board has blind vias, will it significantly affect the price?", this is a more professional and open-ended question. If the similarity score with all entries in the instruction library is below the threshold, the matching fails. At this time, the scheduling logic packages the user input sentence, along with the context of the previous few sentences, and sends it to the configured external AI large language model through the AI ​​interface gateway module. The AI ​​model may reply: "Blind via technology increases manufacturing complexity and the number of laminations, which usually has a certain impact on the price. The specific impact depends on the number of layers and the diameter of the blind vias. To give you an accurate quote, can you tell me the number of layers, dimensions, and specific requirements for the blind vias?" This intelligent reply is returned to the user. At the same time, regardless of the matching result, the information extraction submodule has been activated in this step and begins to work.

[0033] In step S140, the information extraction submodule continues to run during the dialogue processing in step S130. It analyzes all statements in the user's current and historical dialogue. For example, for the user's statement "The board is approximately 10 centimeters long and 5 centimeters wide," the submodule identifies "10 centimeters" and "5 centimeters" as size parameters, performs unit conversion and calculation, and then calculates the area... The area (㎡) is filled into the "Size Area" field of the structured data sheet. For "Use ordinary FR4 material," "FR4" is identified as the material type and filled into the "Material" field. For "Four layers are required," "four layers" is identified and filled into the "Number of Layers" field with a value of 4. For "Immersion gold surface treatment," "immersion gold" is identified and filled into the "Surface Treatment Process Type" field. For "Several locations require impedance control, 90-ohm differential," "impedance control" is identified as a special process, and the parameter "90-ohm differential" is extracted and stored as a special key-value pair {"Special Process_Impedance Control": "90-ohm Differential"} in the data sheet. The data sheet is dynamically maintained in memory and continuously supplemented and corrected as the conversation progresses.

[0034] In step S150, the system continuously monitors the status of the structured data sheet. Pricing trigger conditions are divided into predefined and dynamic judgment types. Predefined conditions are: when the data sheet simultaneously contains the five core parameters—"size / area," "number of layers," "quantity," "material," and "surface treatment process"—the system automatically triggers the pricing process without explicit user instruction. Dynamic judgment conditions are: by analyzing the user's intent, when key intent words such as "quote," "how much," "calculate the price," and "total price" are identified, the system immediately triggers pricing even if the parameters in the data sheet are not yet complete. In this example, suppose the user says after several rounds of dialogue, "Okay, these parameters are almost there, please give me a quote." The system, through intent recognition, determines a clear quotation instruction and immediately triggers pricing. The core pricing engine module is invoked, which reads all valid parameters in the current structured data sheet. Assume the data sheet contains: {"Size / Area": ​​0.005, "Number of Layers": 4, "Quantity": 1, "Material": "FR4", "Surface Treatment": "Immersion Gold", "Special Process_Impedance Control": "90 Ohm Differential"}. The engine first applies the formula based on an area of ​​0.005㎡ (less than 0.05㎡). Calculate the cost of the basic area. Then, calculate the fixed costs of "four-layer board", "FR4 material", and "immersion gold process" in sequence. and area cost For the special process of "impedance control," the engine queries the pricing rule database for the corresponding additional charge rule. Finally, it sums up all the individual charges to arrive at the total price. .

[0035] In step S160, the core pricing engine module calculates the resulting quote. The system returns a brief, optional cost breakdown to the Natural Language Processing and Scheduling module. The Scheduling module converts this structured data into user-friendly, easy-to-understand natural language text. For example: “Based on your requirements (FR4 material, four layers, 50 square centimeters, immersion gold process, including impedance control), the preliminary estimated price is XX yuan. This includes: base board fee YY yuan, layer processing fee ZZ yuan, immersion gold process fee WW yuan, and impedance control surcharge VV yuan. Please note that this is an estimate based on the current parameters; the final price will be based on the confirmed production documents.” This text is presented to the user through the conversational customer service interface of the unified user interface module.

[0036] In step S170, the system temporarily records key node data throughout the entire interaction process from steps S110 to S160. This includes the original dialogue text for each sentence, the system's response text for each instance, the structured data sheet generated by the information extraction submodule, and the input parameters and output results of each pricing request triggered by the core pricing engine module. When the session ends, the system performs final processing: determining the identity status of the interacting user. If the user is an anonymous visitor, these temporary records may be automatically cleared after a certain period of time, or an anonymous session ID may be generated and archived. If the user is logged in, the system strongly binds the complete interaction record to the user's unique account identifier and persistently stores all this data in the backend database through the user account and data recording module, forming an immutable historical record. The record also includes the start and end timestamps of the current session.

[0037] In step S180, logged-in users can access their historical data at any time through the backend data query interface of the unified user interface module. After logging into the backend system, users can see a data query panel. Users can select a time range, such as "October 1, 2023 to October 31, 2023". They can also perform fuzzy searches by entering project keywords, such as searching for all conversations containing the word "RF". More powerfully, users can use process parameters for precise filtering. For example, selecting "Number of Layers" equal to "4" and "Surface Treatment Process" equal to "Immersion Gold" in the filter conditions will list all historical inquiry records that meet the criteria. Clicking on any record allows users to view the complete conversation process, the structured parameters extracted at that time, and detailed quotation calculation results. This data can be exported to Excel or PDF format, facilitating cross-project cost analysis, demand trend summarization, or reporting to internal teams.

[0038] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An AI-based PCB pricing and customer service integration system, characterized in that, The system comprises the following components: a pricing rule database module, used to store PCB price calculation models constructed based on multi-dimensional process factors, including size area, quantity, material type, number of layers, copper thickness, surface treatment process type, number of holes, number of test points, minimum hole diameter, minimum line width and spacing, board utilization rate, rush mark, and other special processes. The pricing logic of the price calculation model is: the sum of the base area cost and the cost of each process factor, where the base area cost is priced in segments according to a preset area range, and the cost of each process factor consists of a fixed surcharge for each item and a surcharge per square meter; a core pricing engine module, used to receive structured process parameter data, call the price calculation model in the pricing rule database module to perform calculations, and output the price results; an AI interface gateway module, providing a configurable application programming interface for accessing at least one external AI large language model; and an AI supplementary instruction library module, storing user-defined AI supplementary instruction pairs, each instruction containing a question and a preset response. The Natural Language Processing and Scheduling module receives and processes the user's natural language input. This module executes the following logic: First, it performs a similarity match between the user's question and the question content in the AI ​​Supplementary Instruction Library module. If the match is successful, it directly returns the corresponding preset response. If the match fails, it sends the user's question and the current conversation context to the configured AI model through the AI ​​Interface Gateway module, and returns the AI ​​model's response to the user. This module also integrates an information extraction submodule, used to identify and extract key process parameters affecting pricing from continuous dialogues, and convert them into structured data that can be processed by the core pricing engine module. The User Account and Data Recording module provides user authentication and login functions, and records the query behavior data of logged-in users. The query behavior data includes original dialogue records, system response content, triggered pricing requests and their results, query timestamps, and query count statistics. The unified user interface module provides a conversational customer service interface and a backend data query interface. The conversational customer service interface supports natural language interaction in the form of a chatbot, and the backend data query interface allows logged-in users to retrieve and analyze their historical interactions and pricing records.

2. The AI-based PCB pricing and customer service integration system according to claim 1, characterized in that, The price calculation model stored in the pricing rule database module has its pricing logic expressed by the following formula: in, Represents the total price. Represents the area of ​​the PCB. Representation based on area The basic area cost function is determined by the interval in which it is located. This represents the total number of technological factors that influence price. Representing the Each process factor incurs a fixed additional charge. Representing the Additional charges per square meter for process factors.

3. The AI-based PCB pricing and customer service integration system according to claim 2, characterized in that, The basic area cost function Defined as a piecewise function, when the area When it is less than 0.05㎡, ;when When the area is in the range of [0.05, 0.1)㎡, ;when When it is greater than or equal to 0.1㎡, ,in , , The preset base price per square meter and .

4. The AI-based PCB pricing and customer service integration system according to claim 1, characterized in that, The similarity matching in the AI ​​supplementary instruction library module is implemented using a cosine similarity algorithm based on word vectors or a semantic similarity calculation based on a pre-trained language model, with a set similarity threshold. When a user's question is similar to any question in the database, the similarity score is calculated. satisfy When the match is successful, it is considered a successful match.

5. The AI-based PCB pricing and customer service integration system according to claim 1, characterized in that, The information extraction submodule in the natural language processing and scheduling module adopts a combination of named entity recognition and relation extraction technology. This submodule has a built-in PCB domain entity dictionary, which scans user statements in real time during the dialogue, identifies and marks entities corresponding to preset process parameter categories, and maintains a dynamically updated structured data sheet through context association in multi-turn dialogues. The data sheet stores parameters in the form of key-value pairs.

6. The AI-based PCB pricing and customer service integration system according to claim 5, characterized in that, The information extraction submodule employs an attention-based sequence labeling model, which is trained on a large amount of labeled PCB requirement dialogue corpus to accurately identify parameter values ​​and their categories in unstructured descriptions.

7. The PCB pricing and customer service integration system based on artificial intelligence according to claim 1, characterized in that, The pricing trigger conditions for the core pricing engine module include predefined conditions and dynamic judgment conditions. The predefined conditions are that pricing is automatically triggered when the structured data sheet contains five core parameters: "size and area", "number of layers", "quantity", "material" and "surface treatment process". The dynamic judgment conditions are that pricing is triggered when the user's intent is analyzed and a preset pricing intent keyword is identified.

8. A method for integrating PCB pricing and customer service based on artificial intelligence, characterized in that, The method includes the following steps: S110, the user initiates an interactive session through the conversational customer service interface of the unified user interface module; S120, the user describes their PCB requirements in natural language; S130, the natural language processing and scheduling module receives the user input and executes the scheduling logic: matching the user input with the AI ​​supplementary instruction library for similarity; if the match is successful, a preset reply is returned; if the match fails, the user input and the current session context are sent to an external AI model through the AI ​​interface gateway to obtain an intelligent reply, and the information extraction submodule is activated; S140, the information extraction submodule runs continuously during the dialogue processing, extracting key process parameters from the user's historical and current statements and filling them into a structured data sheet; S150, when the parameters in the structured data sheet meet the preset pricing trigger conditions, or when the user explicitly issues a quotation instruction, the core pricing engine module is called. This module reads the structured parameters in the data sheet and performs calculations based on the model in the pricing rule database; S160, the system replies to the user in natural language through the conversational customer service interface with the calculated quotation result, along with a description of the cost composition. S170, throughout the entire interaction process, the system temporarily records the original dialogue records, extracted structured parameters, pricing requests and results; if the interacting user is logged in, the complete record is bound to the user account and persistently stored in the user account and data recording module; S180, logged-in users can use the background data query interface of the unified user interface module to search, view and analyze the historical records bound to their accounts based on time range, project keywords or process parameters.

9. The PCB pricing and customer service integration method based on artificial intelligence according to claim 8, characterized in that, The parameter extraction in step S140 employs an attention-based sequence labeling model, which is trained on a large amount of labeled PCB requirement dialogue corpus to accurately identify parameter values ​​and their categories in unstructured descriptions.

10. The PCB pricing and customer service integration method based on artificial intelligence according to claim 8, characterized in that, The pricing triggering conditions in step S150 include predefined conditions and dynamic judgment conditions. The predefined conditions are that pricing is automatically triggered when the structured data sheet contains the five core parameters of "size area", "number of layers", "quantity", "material" and "surface treatment process". The dynamic judgment conditions are that pricing is triggered when the user's intent is analyzed and a preset pricing intent keyword is identified.

Images