At present, the research units of the oil pump industry focus on the basic research of injection mechanism, test mechanism and so on. The production unit focuses on product imitation and modification, and seldom studies the design and development process of oil pump products. Most domestic manufacturers of new products from design to stereotyping experienced a long time, resulting in long product development cycle, poor quality. MechNet | For more information, please visit the China Machinery Expert Network. In view of the shortcomings in the product development process of the oil pump production industry, we propose a design method of oil pump based on knowledge and visualization technology (Figure 1). By improving and perfecting the design process of oil pump products, this method can make the product development process model cover user needs, preliminary design, injection performance analysis, detailed design and assembly analysis. MechNet | More information, please visit the China Machinery expert network, can be seen from Figure 1, in the design stage, you can use the instance-based design method or model-based design method for the design of the main key parameters. The former will make full use of the program design case base, retrieve the program according to the appropriate retrieval method and change the instances according to the demand changes; the latter design the different empirical coefficients based on the mathematical model. After the program design, the injection performance simulation can be carried out, and the modification of some main parameters is decided by the performance evaluation and redesign system. Detailed design, mainly through the constraint-based product family design method, the structural design of all components and all parts of the parameter design, you can perform assembly simulation analysis. Various requirements to meet, you can get more satisfactory digital products optimized pump program. Application of simulation technology and artificial intelligence technology, will make product development to the height of the "virtual product development", and in the design phase can consider the design of many activities in the post-event, thereby shortening the product development cycle and improve the test success rate. Figure 1 based on knowledge and visualization technology of the pump design process 2 based on the example of the pump design From the production of oil pump nozzle in the 1950s, has gone from a copy to the modified technology development path, which is characterized by full use of design experience and examples , Case-based design (Case Besed Design, CBD) is a suitable method for oil pump development. The design process is based on design tasks, selected from the prototype prototype instance of the prototype and generated to the current work area [1,2]. Instance Prototypes Retrieve similar instances from instance space based on their instance retrieval model. If a similar instance is found, it is extracted and mapped to a similar target scenario; if it is verified that the target scenario fully satisfies the current design task requirements, the instance can serve as the final design match the design task and store in the instance repository; if not, Design task requirements, then analyze it and modify it until all design requirements are met. Thus, the design of instance-based is divided into instance expression, similar instance retrieval, instance modification, instance verification and so on. The design example consists of the data of the design instance, the solution knowledge of the instance and the index of the instance. Oil pump design uses object-oriented knowledge representation method, that is, a variety of single knowledge expression methods (rules, frameworks and processes), in accordance with the principles of object-oriented design to form a mixed knowledge representation. With the pump as the center, the static attributes and dynamic behavior characteristics, as well as the design process knowledge, are "encapsulated" in the structure of the presentation object. This method allows the complex oil pump object to be decomposed into several simple objects and becomes a tree structure (as shown in FIG. 2). At the same time, a complex example can be gradually broken down into simpler examples (Figure 3). Fig. 2 Spraying system tree Fig. 3 Decomposition structure of the example There are two kinds of decomposing structure degree: â‘ If the detailed design is supported, it can be decomposed to the part level, but it may cause the structure of the instance library to be huge and complex; â‘¡ If the supporting scheme is designed, Can carry out the program design evaluation (injection performance evaluation) of the structural characteristics and the main key components feature parameters level. The extraction of the best example is to extract in the instance library the instance or instance fragment whose characteristics are most similar to the design task characteristics. Design tasks include design goals, constraints, and initial conditions. The search for similar instances is actually a decision in multi-attribute space, so the weighted sum based on index features is a typical search strategy. When an instance is represented as a collection, the distance between instances can be defined as the intersection of two instances; when an instance is represented as a state space vector, the distance between instances can be defined as the distance between vectors on the algebra. All three belong to the nearest neighbor search strategy. When the design examples have enough time, can also be based on neural network extraction strategy. The neural network-based extraction strategy can eliminate the interference of human factors through the training of multiple successful and failed instances, and has the ability of self-adaptation and self-learning. 3 Design knowledge from design process data When the similarity between design examples is low, the design method of CBD does not apply and the mapping between design, performance, structure and parameters reflects the design experience and is not conducive to design experience The accumulation and dissemination. Product family-based design reflects design knowledge from constraints and can not guide the design process. Therefore, it is necessary to seek a scientific calculation model and simulation model to improve the product's self-development capability. Due to the particularity of diesel engine, its performance is good or not depends on the match between fuel injection, intake air and combustion, so there are many empirical formulas and empirical coefficients for design calculation model and injection performance model. Here, we use the knowledge discovery in database (KDD) technology principle to seek product characteristics, design intent and design knowledge from the previous product instance library and test conclusion library. After years of extensive application of computer aided design, production, management and database technology, product development department has recorded a large number of database related to product development, such as design task feature library, product list and pump commissioning evaluation documents. In order to transform the latent knowledge in the design process into explicit knowledge, the author uses the database knowledge discovery technology to excavate the design knowledge. KDD is a technology that obtains correct, novel, potentially useful and finally understandable patterns from the database [3]. There are many empirical coefficients in the pump design and calculation model. The selection of these factors is directly related to the performance of the product and the scale of the design. Using the design task feature library, product specification, commissioning evaluation results and design calculation model, we recalculate all kinds of past design examples to get the design task feature library, the design process experience coefficient library corresponding to the task features, the main key parameter database and Design program evaluation conclusion library. KDD method through the design knowledge mining, oil pump design knowledge (Figure 4). Figure 4 injection pump design knowledge mining principle Pump design data mining steps: (1) select and prepare the data to be excavated. According to the existing product design requirements, the main key parameters and test conclusions, the design experience coefficient of each product instance is inverted; (2) Data preprocessing. Establish a relational database to describe the product design requirements, the main key parameters and test conclusions, the other to build a database to describe the design experience coefficient of the corresponding product instance, to reduce the complexity of the data and to reorganize the data through purification, reduction, transformation, classification;