【正文】 sistency andintegrity of the decision space within the given constraints. The product design data are sent to theCoCAPP system through the Dagent from the puteraided design system. The CoCAPP system getsFig. 1. The structure of the Machining CoCAPP system.(). Zhao et in Industry 41 2020 83–9786the production constraints from the schedulingrshopfloor system. The CoCAPP system generates theprocess plans according to the product design dataand production constraints. If no acceptable resultscan be obtained, the CoCAPP system will feed theconflict information back to the design department orother relevant departments. The feasible process planalternatives will eventually be transmitted to theschedulingrshopfloor system for scheduling.. Agent infrastructureThe general structure of process planning agents .Pagents is shown in Fig. 2. The agent is posedof four parts: agent controller, inference engine,functional adapters, and application libraries. Theconfiguration file is used to construct the agent. Therules and facts form the application libraries. Theagent controller, rulebased engine, inter adapter,file adapter, keyboard adapter, information viewadapter, and schedule adapter are the monly usedponents for all the Pagents. Other adapters arealso shown in Fig. 2.The solver adapter is a very important adapter inthe Pagents because it is used to acplish proposal generation, conflict resolution and proposalevaluation of the process planning. In order to utilizethe knowledge of each domain in the bination ofproduction rules and objects, the adapter is differently implemented for different agents. It must bespecifically designed to deal with the knowledge ofobjectoriented description.The database adapter is used to store data usefulto the Bagent and Pagents, such as problem definitions, proposals, conflicts, evaluations, solutions, etc.wxBecause KQML 17 is the most monly usedlanguage for munication among agentbased programs, particularly when they are autonomous andasynchronous, the CoCAPP system has chosen aKQMLbased munication protocol as a munication language used by each agent. The NetKQMLadapter is used to municate with the Bagent. .Fig. 2. The structure of process planning agents Pagents .(). Zhao et in Industry 41 2020 83–97 87According to the requirements of KQML transport,agents are connected by unidirectional munication links that carry discrete messages. These linksmay have a finite message transport delay associatedwith them. When an agent receives a message, itknows from which ining link the message hasarrived。fl cutting tool selection agent。fl machining operation selection agent。fl fixture selection and design。fl machine selection。 etc. In the past, process plans were often generatedby human process planners who had plenty of manufacturing domain knowledge and worthy experience.In the recent decades, puter technologies havestimulated the advance toward puteraided pro .cess planning CAPP .Generally, there are two CAPP approaches: variant and generative. The variant approach is a dataretrieval and editing method. Some standard or mature process plans are collected based on the grouptechnology and stored in a database. When a newpart is required to be produced, a similar processplan is retrieved from the database and edited toadjust it to suit the new part. The generative approach is a knowledgebased method which automat01663615r00r$ see front matter q 2020 Elsevier Science . All rights reserved. .PII: S01663615 99 000123(). Zhao et in Industry 41 2020 83–9784ically generates a process plan according to thepart’s features and manufacturing requirements.The success of the variant approach depends onthe group technology, planner’s experience and asufficient collection of standard or mature processplans. This method is especially suitable for panies with few product families and a large number ofparts per family. Most earlier CAPP tools can becategorized under the variant process planning apwx wxproach 1 . Typical examples are CAPP 2 , MIwxPLAN 3 , etc. The generative process planning approach has attracted more attention in recent years. Itoffers a potential of producing an optimal plan.wx wxTypical examples are APPAS 2 EXCAP 4 , KROwx wx .NOS 5 , XCUT 6 , QTC Quick turnaround cellwx wx 2 , PART 7 , OOPPS objectoriented process plan. wx wxning system 8 , MePlans 9 , COMPLAN Process .wxPlanner CPP 10 , etc. Generative process planningsystems are mostly oriented towards the needs oflarge panies and research organizations, especially those which have a number of products insmall lot sizes. However, there is still difficulty indeveloping a truly generative process planning system which can meet industrial needs and provide anappropriate and patible generic framework,knowledge representation method, and inferencemechanism.Cooperative agent systems attempt to distributeactivities to multiple specialized problem solvers andwto coordinate them to solve plex problems 11–x14 . A cooperative agent system consists of manyindividual agents with cooperation engines. Eachagent which has its own knowledge base and inference engine is responsible for a specific task. Itprovides a cooperation interface to municate withother agents in the cooperative environment. A different language and different knowledge representation may be employed by each agent which may wellbe located in a different machine. Such a systemorganization provides an integration environment ofheterogeneous and scalable architecture suitable tosolving different problems.2. Process planning problemA machining process generally involves manymachine tools, operations, fixtures, and cutting tools.Its planning requires k