WO2022126564A1 - 高可靠抗弧光干扰的焊接多形态数据智能体及处理方法 - Google Patents

高可靠抗弧光干扰的焊接多形态数据智能体及处理方法 Download PDF

Info

Publication number
WO2022126564A1
WO2022126564A1 PCT/CN2020/137415 CN2020137415W WO2022126564A1 WO 2022126564 A1 WO2022126564 A1 WO 2022126564A1 CN 2020137415 W CN2020137415 W CN 2020137415W WO 2022126564 A1 WO2022126564 A1 WO 2022126564A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
intelligent
welding
layer
physical
Prior art date
Application number
PCT/CN2020/137415
Other languages
English (en)
French (fr)
Inventor
冯消冰
刘文龙
Original Assignee
南京水木自动化科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 南京水木自动化科技有限公司 filed Critical 南京水木自动化科技有限公司
Priority to PCT/CN2020/137415 priority Critical patent/WO2022126564A1/zh
Publication of WO2022126564A1 publication Critical patent/WO2022126564A1/zh

Links

Classifications

    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B23—MACHINE TOOLS; METAL-WORKING NOT OTHERWISE PROVIDED FOR
    • B23K—SOLDERING OR UNSOLDERING; WELDING; CLADDING OR PLATING BY SOLDERING OR WELDING; CUTTING BY APPLYING HEAT LOCALLY, e.g. FLAME CUTTING; WORKING BY LASER BEAM
    • B23K37/00—Auxiliary devices or processes, not specially adapted to a procedure covered by only one of the preceding main groups

Definitions

  • the invention relates to the technical field of welding robots, in particular to a highly reliable and anti-arc interference welding polymorphic data intelligent body and a processing method.
  • Welding robots are industrial robots engaged in welding, and have had an important impact on various fields of modern high-tech industries and even people's lives.
  • the main body of the robot travels to the surface of a large steel structure for welding, and is located on the ground or scaffolding.
  • the control device on the robot controls the main body of the robot to complete the welding work.
  • the inventor found that the data generated in the welding process can be used to feedback the product quality defects in the welding robot, and the performance optimization of the welding robot can be fed back through the data to further ensure the welding quality, and finally optimize and iterate a high-quality welding robot.
  • the data generated in the welding process can be used to feedback the product quality defects in the welding robot, and the performance optimization of the welding robot can be fed back through the data to further ensure the welding quality, and finally optimize and iterate a high-quality welding robot.
  • how to manage them effectively is a complicated problem.
  • the purpose of the present invention is to provide a highly reliable anti-arc interference welding polymorphic data agent and processing method, which can process the original welding data of various forms to form usable data that is conducive to analysis, thereby abstracting product quality. question.
  • the present invention provides a welding polymorphic data intelligent body with high reliability and anti-arc interference, including: a main controller, which is used to collect data in real time from various sensing devices of the crawling welding robot, and provides Multiple data interfaces; communication controller, which is connected to the cloud status monitoring device; physical anti-interference isolation layer, which is set between the main controller and the communication controller, and the physical anti-interference isolation layer is used for outer isolation; application layer , which is communicatively connected to the main controller, and the application layer includes an application-layer intelligent master node and an application-layer intelligent standby node that are connected in communication, wherein, when the application-layer intelligent master node fails, the application-layer intelligent master node and the application-layer intelligent standby node State switching is possible, and both the application layer intelligent master node and the application layer intelligent standby node include a composite intelligent body module with a plurality of intelligent algorithms; and a physical entity storage layer, which is communicatively connected to the application layer, and the physical entity storage layer
  • the state switch in the process of entering the cloud synchronization, is changed to the off state. If the crawling welding robot is in the running state at this time, the time consistency verification algorithm will logically verify the normalized morphological data. It is sent to the slice stack backup queue; when the cloud synchronization is completed, the slice stack backup queue is merged into the slice stack queue, and the status switch is changed to the ON state.
  • transforming the image data of the weld through the image feature characterization algorithm includes the following steps: first, the RGB image is subjected to binarization grayscale processing, and then scanned through a plurality of neural nuclei, and the convolution calculation is performed. Low-dimensional data, through multiple low-dimensional processing, finally output several 1*N-dimensional data, and then intercept different segments to represent each data, so that effective features can be extracted from the weld image according to the time dimension and represented as structured multi-dimensional data .
  • the abnormal screening and processing of the original data of the welding machine data and the main push data includes the following steps: filtering the abnormal time series data on the original data of the welding machine data and the main push data, and then finding out all problematic data. A collection of time points, and all data in this collection are eliminated to obtain normalized morphological data.
  • the data collected by the main controller includes image data, welding machine data and machine body motion data.
  • the present invention configures the application layer to include an application layer intelligent master node and an application layer intelligent backup node.
  • Active and standby dual node management the physical entity storage layer is set to include the physical layer storage entity master node and the physical layer storage entity standby node.
  • the welding data is not lost to the greatest extent, and the welding quality data, welding machine data and robot body motion data generated by welding are recorded.
  • Figure 1 is the structure of the highly reliable anti-arc interference welding polymorphic data agent of the preferred embodiment of the present invention
  • FIG. 2 is a flow chart of a method for intelligent processing of welding polymorphic data with high reliability and anti-arc interference according to a preferred embodiment of the present invention.
  • the highly reliable anti-arc interference welding polymorphic data agent includes: a main controller 101, a communication controller 102, a physical anti-interference isolation layer 103, and an application layer 104 and the physical physical storage layer 105.
  • the main controller 101 is used to collect data in real time from various sensing devices of the crawling welding robot 100, and provides multiple data interfaces.
  • the communication controller 102 is in communication connection with the cloud state monitoring device 106 .
  • the physical anti-interference isolation layer 103 is arranged between the main controller 101 and the communication controller 102, and the physical anti-interference isolation layer 103 is used for outer layer isolation.
  • the application layer 104 is in communication connection with the main controller 101, and the application layer 104 includes an application layer intelligent master node 114 and an application layer intelligent standby node 124 that are communicatively connected.
  • the application layer intelligent master node 114 fails, the application layer intelligent master node 114 and the application layer intelligent standby node 124 can perform state switching, thereby quickly realizing the time series data processing and response capabilities of the agent.
  • the application layer intelligent master node 114 and the application layer intelligent standby node 124 each include a composite agent module 134 having a plurality of intelligent algorithms.
  • the physical entity storage layer 105 is in communication connection with the application layer 104.
  • the physical entity storage layer 105 includes a physical layer storage entity master node 115 and a physical layer storage entity backup node 125 that are communicatively connected, and the physical layer storage entity master node 115 and the application layer intelligent master node. 114 Communication connection. Wherein, when the physical layer storage entity master node 115 fails, the physical layer storage entity master node 115 and the physical layer storage entity backup node 125 can make the physical entity storage layer 105 work normally through state switching.
  • a composite agent with multiple intelligent algorithms includes a polymorphic time series data queue, an image feature representation intelligent algorithm body, an anomaly detection intelligent algorithm body, a time check consistency intelligent algorithm body, and an active and standby slice heap processing. body.
  • the physical entity storage layer includes an I/O read/write function and a storage function, and the I/O read/write function requires that the read/write speed is not lower than the frequency of commands issued by the main controller.
  • the storage function satisfies the ability to report twice the failure period in the extreme case of data failure in the cloud, that is, the locally reserved free storage space resources must be guaranteed by 2 times.
  • the physical anti-interference isolation layer is a carbon nanotube (CNT) fiber material, which is used to absorb the electromagnetic radiation of the arc light.
  • CNT carbon nanotube
  • the data collected by the main controller includes image data, welding machine data and machine body motion data
  • the multiple data interfaces include a time series image data interface, a main push time series data interface, a welding time series data interface and a slider time series.
  • the data interface is used to transmit time series images, main push data, welding machine data and slider data respectively.
  • the present invention also discloses a highly reliable anti-arc interference intelligent processing method for welding polymorphic data, including the following steps: constructing a main controller 101, a physical anti-interference isolation layer 103, an application layer 104, a physical entity The storage layer 105 and the communication controller 102 ; the main controller 101 collects data in real time from multiple sensing devices of the crawling welding robot 100 .
  • the main controller 101 transmits the collected data to the composite intelligent body module with multiple intelligent algorithms of the intelligent master node of the application layer through multiple data interfaces.
  • the multiple data interfaces include a time series image data interface, a main push time series data interface, a welding time series data interface and a slider time series data interface, which are respectively used to transmit time series images 201, main push data 202, welding machine data 203 and slider data 204.
  • the composite agent module When the composite agent module receives the time series image 201, the main push data 202, the welding machine data 203 and the slider data 204, the time series image 201, the main push data 202, the welding machine data 203 and the slider data 204 enter the time series queue 211.
  • the image feature characterization intelligent algorithm body 212 converts the image data of the weld through the image feature characterization algorithm, outputs the parameters of the unwelded weld, and outputs the welding quality data that has been welded.
  • the abnormality detection intelligent algorithm body 213 performs abnormality discrimination on the transformed data through the abnormality detection intelligent algorithm, and performs abnormality discrimination and processing on the welding machine data and the original data of the main push data, and obtains normalized morphological data.
  • the temporal consistency checking algorithm body 214 performs logical checking on the normalized morphological data through the temporal consistency checking algorithm, and obtains time-series single morphological data.
  • the time series single morphological data is subjected to heap processing to form extractable time axis playback data.
  • the time axis playback data is subjected to slice stack processing in the slice stack queue 215 and stored in the physical entity storage layer. 105.
  • the cloud state monitoring device 106 periodically obtains the data stored in the physical entity storage layer 105 through the communication controller 102 synchronously.
  • both the application layer and the physical entity storage layer have the dual-node disaster recovery and backup switching function of the active and standby nodes.
  • the application layer intelligent master node fails, the application layer intelligent master node and the application layer intelligent standby node perform state switching; and when the physical layer storage entity master node fails, the physical layer storage entity master node and the physical layer storage entity standby node.
  • the physical physical storage layer works normally through state switching.
  • the state switch 217 in the process of entering the cloud synchronization, is changed to the off state. If the crawling welding robot 100 is in the running state at this time, the time consistency check algorithm normalizes the logic check.
  • the morphological data is sent to the slice stack backup queue 216; when the cloud synchronization is completed, the slice stack backup queue 216 is merged into the slice stack queue 215, and the state switch 217 is changed to the ON state.
  • transforming the image data of the weld through an image feature characterization algorithm includes:
  • the following steps first perform binary grayscale processing on the RGB image, then scan through multiple neural nuclei, calculate low-dimensional data by convolution, and finally output several 1*N-dimensional data through multiple low-dimensional processing, and then separately Different segments are intercepted to represent each data, so that effective features can be extracted from the weld image according to the time dimension and represented as structured multi-dimensional data.
  • abnormal screening and processing are performed on the original data of the welding machine data and the main push data to obtain normalized morphological data, including the following steps: the original data of the welding machine data and the main push data are processed as abnormal time series data. Filter, and then find out all problematic time point sets, and remove all data in this set to obtain normalized morphological data.
  • the welding machine data (such as current and voltage) has a certain reasonable value range, but when it is transmitted through the controller, due to various reasons, there will be abnormal conditions, which exceed reasonable expectations, and obvious error data will appear, so abnormal
  • the detection algorithm will ensure that reasonable data is retained and unreasonable data is deleted through several means:
  • the values of two adjacent time points vibrate violently, with large-span jumps, and they will also be deleted if they exceed a reasonable range. 3.
  • the value of the continuous time period is close to the reasonable upper and lower thresholds. It is necessary to add logical judgment and conditional detection and deletion.
  • the normalized morphological data means that the types of data from different welding machine bodies are inconsistent, and the upper and lower limit value ranges are different, which needs to be unified here. Flatten to 0 ⁇ 1 range, such as voltage/current/main thrust speed, etc.
  • the time consistency check algorithm is based on the data in the interval 0 ⁇ 1 at all the above time points. For example, the result from the image algorithm at 0:01 shows that welding is in progress, but the speed of the main pusher is normal. The normalization value is close to 0. It may be that the time information returned by a certain welding machine component is incorrect, resulting in a time misalignment problem. Therefore, it is necessary to perform time correction again and do a feedback process.
  • the output of the time consistency check algorithm is managed separately according to the dimensions of the different bodies of the welding machine robot.
  • the welding machine has a set of 0 ⁇ 1 data
  • the main pusher has a set of 0 ⁇ 1 data
  • the slider has a set of 0 ⁇ 1 data, all of which are in chronological order.
  • the heap processing is sorted according to the time dimension and regularly compressed (such as every minute) to form a bag. document.
  • the invention ensures the effective management and feedback of the polymorphic data of tens of millions of welding robots in end-testing through the excellent advanced technology of software and hardware, and finally can form the product portrait of the real product of the welding robot, which is used to guide the R&D iteration and precise optimization of the robot.
  • the embodiments of the present application may be provided as a method, a system, or a computer program product.
  • the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects.
  • the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

Landscapes

  • Physics & Mathematics (AREA)
  • Optics & Photonics (AREA)
  • Engineering & Computer Science (AREA)
  • Mechanical Engineering (AREA)
  • Manipulator (AREA)
  • Numerical Control (AREA)

Abstract

一种高可靠抗弧光干扰的焊接多形态数据智能体及处理方法，包括：主控制器（101），其用于从爬行焊接机器人（100）的各类传感设备上实时采集数据；通讯控制器（102），其与云端状态监听设备（106）通信连接；物理抗干扰隔离层（103），其设置于主控制器（101）与通讯控制器（102）之间，物理抗干扰隔离层（103）用于外层隔离；应用层（104），其与主控制器（101）通信连接，应用层（104）包括通信连接的应用层智能主节点（114）和应用层智能备用节点（124）；以及物理实体存储层（105），其与应用层（104）通信连接，物理实体存储层（105）包括通信连接的物理层存储实体主节点（115）和物理层存储实体备用节点（125），物理层存储实体主节点（115）与应用层智能主节点（114）通信连接。

Description

高可靠抗弧光干扰的焊接多形态数据智能体及处理方法 技术领域
本发明涉及焊接机器人技术领域，具体涉及一种高可靠抗弧光干扰的焊接多形态数据智能体及处理方法。
 背景技术
焊接机器人是从事焊接的工业机器人，并且对现代高技术产业各领域以至人们的生活产生了重要影响，焊接机器人在工作状态时，机器人主体行进至大型钢结构体表面进行焊接，由位于地面或脚手架上的控制设备对机器人主体进行控制，以完成焊接工作。发明人发现可以通过焊接过程中产生的数据，反馈出在焊接机器人的产品质量缺陷，通过数据反哺焊接机器人的性能优化，进一步保证焊接质量，最终优化迭代出高质量的焊接机器人。但是由于焊接过程中，涉及的数据种类较多，如何有效管理，是一个比较复杂的难题。公开于该背景技术部分的信息仅仅旨在增加对本发明的总体背景的理解，而不应当被视为承认或以任何形式暗示该信息构成已为本领域一般技术人员所公知的现有技术。公开于该背景技术部分的信息仅仅旨在增加对本发明的总体背景的理解，而不应当被视为承认或以任何形式暗示该信息构成已为本领域一般技术人员所公知的现有技术。
 技术问题
本发明的目的在于提供一种高可靠抗弧光干扰的焊接多形态数据智能体及处理方法，其能够通过将各种形态的原始焊接数据，加工形成利于分析的可利用数据，从而抽象出 产品质量问题。
 技术解决方案
为实现上述目的，本发明提供了一种高可靠抗弧光干扰的焊接多形态数据智能体，包括：主控制器，其用于从爬行焊接机器人的各类传感设备上实时采集数据，并提供多个数据接口；通讯控制器，其与云端状态监听设备通信连接；物理抗干扰隔离层，其设置于主控制器与通讯控制器之间，物理抗干扰隔离层用于外层隔离；应用层，其与主控制器通信连接，应用层包括通信连接的应用层智能主节点和应用层智能备用节点，其中，当应用层智能主节点发生故障时，应用层智能主节点和应用层智能备用节点能够进行状态倒换，并且应用层智能主节点和应用层智能备用节点均包括具有多个智能算法的复合智能体模块；以及物理实体存储层，其与应用层通信连接，物理实体存储层包括通信连接的物理层存储实体主节点和物理层存储实体备用节点，物理层存储实体主节点与应用层智能主节点通信连接，其中，当物理层存储实体主节点发生故障时，物理层存储实体主节点和物理层存储实体备用节点能够通过状态倒换来使物理实体存储层正常工作。
在一优选实施方式中，在进入云端同步的过程中，状态开关改为关闭状态，如果此时爬行焊接机器人处于运行态状态，时间一致性校验算法把逻辑校验后的归一化形态数据输送到切片堆栈备用队列；当云端同步结束之后，切片堆栈备用队列归并到切片堆栈队列，状态开关改为开启状态。
在一优选实施方式中，通过图像特征表征算法将焊缝的图像数据进行转化处理包括如下步骤：先对 RGB 图像做二值化灰度处理，然后通过多个神经核去扫描，卷积计算出低维数据，通过多次低维度处理，最终输出若干 1*N 维数据，然后分别截取不同段表示各个数据，从而能够按照时间维度将焊缝图像提取出有效特征，并表征为结构化多维数据。
在一优选实施方式中，对焊机数据和主推数据的原始数据做异常甄别和处理包括如下步骤：将焊机数据和主推数据的原始数据做异常时序数据的过滤，然后找出所有有问题的时间点集合，把该集合下所有数据全部做剔除处理，以得到归一化形态数据。
在一优选实施方式中，主控制器所采集的数据包括图像数据、焊机数据和机器本体运动数据。
 有益效果
与现有技术相比，本发明的高可靠抗弧光干扰的焊接多形态数据智能体及处理方法的有益效果如下：本发明通过将应用层设置为包括应用层智能主节点和应用层智能备用节点的主备双节点管理、将物理实体存储层设置为包括物理层存储实体主节点和物理层存储实体备用节点主备双节点管理，使得本发明的焊接多形态数据智能体具有高可靠性，可以让焊接数据最大程度上不丢失，记录焊接产生的焊接质量数据、焊机数据和机器人主体运动数据。
通过设置多种人工智能算法组合的复合智能体，有效保障了多形态机器人运行态数据的可阅读性与可利用性，打通了焊接数据集中化管理与处理的瓶颈；通过设置物理抗干扰隔离层，可以有效屏蔽弧光以及特殊频段电磁波，从而保障了焊接时序数据的真实性。通过设置物理抗干扰隔离层可以避免弧光以及部分电磁场的干扰，保证原生态数据高质量的传送到云端，为真实反应千万级规模量的机器人运行态奠定可信任的硬件底座框架。
 附图说明
图 1 为本发明的优选实施方式的高可靠抗弧光干扰的焊接多形态数据智能体的结构
框图。
图 2 为本发明的优选实施方式的高可靠抗弧光干扰的焊接多形态数据智能处理方法流程框图。
 本发明的实施方式
下面将结合附图对本发明实施例中的技术方案进行清楚、完整地描述。本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。
除非另有其它明确表示，否则在整个说明书和权利要求书中，术语“包括”或其变换如“包含”或“包括有”等等将被理解为包括所陈述的元件或组成部分，而并未排除其它元件或其它组成部分。
实施例 1
如图1-2所示，根据本发明一优选实施方式的高可靠抗弧光干扰的焊接多形态数据智能体，包括：主控制器 101、通讯控制器 102、物理抗干扰隔离层 103、应用层 104 以及物理实体存储层 105。其中，主控制器 101 用于从爬行焊接机器人 100 的各类传感设备上实时采集数据，并提供多个数据接口。通讯控制器 102 与云端状态监听设备 106 通信连接。物理抗干扰隔离层 103 设置于主控制器 101 与通讯控制器 102 之间，物理抗干扰隔离层 103 用于外层隔离。应用层 104 与主控制器 101 通信连接，应用层 104 包括通信连接的应用层智能主节点 114 和应用层智能备用节点 124。当应用层智能主节点 114 发生故障时，应用层智能主节点 114 和应用层智能备用节点 124 能够进行状态倒换，从而快速实现智能体的时序数据处理和响应能力。应用层智能主节点 114 和应用层智能备用节点 124 均包括具有多个智能算法的复合智能体模块 134。物理实体存储层 105 与应用层 104 通信连接，物理实体存储层 105包括通信连接的物理层存储实体主节点 115 和物理层存储实体备用节点 125，物理层存储实体主节点 115 与应用层智能主节点 114 通信连接。其中，当物理层存储实体主节点 115 发生故障时，物理层存储实体主节点 115 和物理层存储实体备用节点 125 能够通过状态倒换来使物理实体存储层 105 正常工作。
在一优选实施方式中，具有多个智能算法的复合智能体包括多形态时序数据队列，图像特征表征智能算法体、异常检测智能算法体、时间校验一致性智能算法体和主备切片堆处理体。
在一优选实施方式中，物理实体存储层包括I/O读写功能和存储功能，I/O读写功能要求读写速度不低于主控制器下发指令的频率。存储功能满足在云端上报数据失效极端情况下失效周期的 2 倍能力，即本地预留的 free 存储空间资源要按照 2 倍来保证。
在一优选实施方式中，物理抗干扰隔离层为碳纳米管(CNT)纤维材料，其用于吸收弧光的电磁辐射。
在一优选实施方式中，主控制器所采集的数据包括图像数据、焊机数据和机器本体运动数据，多个数据接口包括时序图像数据接口、主推时序数据接口、焊接时序数据接口和滑块时序数据接口，分别用于传输时序图像、主推数据、焊机数据和滑块数据。
实施例 2
如图2所示，本发明还公开了一种高可靠抗弧光干扰的焊接多形态数据智能处理方法，包括如下步骤：构建主控制器101、物理抗干扰隔离层103、应用层104、物理实体存储层105以及通讯控制器 102；主控制器 101 从爬行焊接机器人 100 的多个传感设备实时采集数据。
主控器101通过多个数据接口将所采集的数据传输给应用层智能主节点的具有多个智能算法的复合智能体模块。其中，多个数据接口包括时序图像数据接口、主推时序数据接口、焊接时序数据接口和滑块时序数据接口，分别用于传输时序图像 201、主推数据 202、焊机数据 203 和滑块数据 204。
当复合智能体模块所接收到的时序图像201、主推数据202、焊机数据203和滑块数据 204 时，时序图像 201、主推数据 202、焊机数据 203 和滑块数据 204 进入时序队列211。
图像特征表征智能算法体212通过图像特征表征算法将焊缝的图像数据进行转化处理，输出未焊接的焊缝参数，并输出已经焊接完成的焊接质量数据。
异常检测智能算法体213通过异常检测智能算法对转化处理后的数据进行异常甄别，并且对焊机数据和主推数据的原始数据做异常甄别和处理，并得到归一化形态数据。
时间一致性校验算法体214通过时间一致性校验算法将归一化形态数据进行逻辑校验，并得到时序单一形态数据。
在状态开关217处于开启的状态下，把时序单一形态数据进行堆处理，形成可提取的时间轴回放数据，时间轴回放数据在切片堆栈队列 215 中进行切片堆栈处理，并存储在物理实体存储层 105 中。
云端状态监听设备106通过通讯控制器102定时同步获取存储在物理实体存储层105中的数据。其中，应用层和物理实体存储层均具有主备双节点容灾备份倒换功能。当应用层智能主节点发生故障时，应用层智能主节点和应用层智能备用节点进行状态倒换；并且当物理层存储实体主节点发生故障时，物理层存储实体主节点和物理层存储实体备用节点通过状态倒换来使物理实体存储层正常工作。
实施例 3
在一优选实施方式中，在进入云端同步的过程中，状态开关217改为关闭状态，如果此时爬行焊接机器人 100 处于运行态状态，时间一致性校验算法把逻辑校验后的归一化形态数据输送到切片堆栈备用队列 216；当云端同步结束之后，切片堆栈备用队列 216 归并到切片堆栈队列 215，状态开关 217 改为开启状态。
在一优选实施方式中，通过图像特征表征算法将焊缝的图像数据进行转化处理包括
如下步骤：先对 RGB 图像做二值化灰度处理，然后通过多个神经核去扫描，卷积计算出低维数据，通过多次低维度处理，最终输出若干 1*N 维数据，然后分别截取不同段表示各个数据，从而能够按照时间维度将焊缝图像提取出有效特征，并表征为结构化多维数据。
在一优选实施方式中，对焊机数据和主推数据的原始数据做异常甄别和处理，以得到归一化形态数据，包括如下步骤：将焊机数据和主推数据的原始数据做异常时序数据的过滤，然后找出所有有问题的时间点集合，把该集合下所有数据全部做剔除处理，以得到归一化形态数据。例如，焊机数据（如电流电压）是有一定合理取值范围，但是在经过控制器传送过来的时候，由于各种原因，会出现异常情况，超出合理预期，出现明显的错误数据，因此异常检测算法会通过几种手段保证合理的数据予以保留，不合理的数据予以删除：
1.  预先设定取值范围如[100，250]，超过就删除。
2.两个相邻时间点的取值发生剧烈抖动，大跨度跳动，超过合理范围也会删除。3.连续时间段取值接近合理上下限阈值，需要增加逻辑判断，条件性检测删除，归一化形态数据是指不同焊机主体来的数据种类不一致，上下限取值范围不同，这里需要统一拉平到 0~1 区间，比如电压/电流/主推速度等。
在一优选实施方式中，时间一致性校验算法是在上面所有时间点上0~1区间的数据进行判断，比如 0：01 时刻图像算法出来的结果显示正在焊接中，但是主推机器的速度归一化值接近 0，有可能是某个焊机组件返回的时间信息不对，产生了时间错位问题。因此需要重新进行时间修正，并且做一个反馈处理。时间一致性校验算法的输出是按照焊机机器人不同主体的维度来分别管理。如焊机有一组 0~1 数据，主推有一组 0~1 数据，滑块有一组 0~1数据，全部有时间顺序，堆处理是按照时间维度排序定期压缩处理（如每分钟），形成 bag文件。
本发明通过过硬的软硬件先进技术，保证了端测千万台焊接机器人的多态数据有效管理与反馈，最终可形成焊接机器人真实产品的产品画像，用于指导机器人的研发迭代与精准优化。
本领域内的技术人员应明白，本申请的实施例可提供为方法、***、或计算机程序产品。
因此，本申请可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且，本申请可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质（包括但不限于磁盘存储器、CD-ROM、光学存储器等）上实施的计算机程序产品的形式。
尽管已经示出和描述了本发明的实施例，对于本领域的普通技术人员而言，可以理解在不脱离本发明的原理和精神的情况下可以对这些实施例进行多种变化、修改、替换和变型，本发明的范围由所附权利要求及其等同物限定。

Claims (10)

  1. 一种高可靠抗弧光干扰的焊接多形态数据智能体，其特征在于：所述焊接多形态数据智能体包括：主控制器，其用于从爬行焊接机器人的各类传感设备上实时采集数据，并提供多个数据接口；
    通讯控制器，其与云端状态监听设备通信连接；物理抗干扰隔离层，其设置于所述主控制器与通讯控制器之间，所述物理抗干扰隔离层用于外层隔离；应用层，其与所述主控制器通信连接，所述应用层包括通信连接的应用层智能主节点和应用层智能备用节点，其中，当所述应用层智能主节点发生故障时，所述应用层智能主节点和应用层智能备用节点能够进行状态倒换，并且所述应用层智能主节点和所述应用层智能备用节点均包括具有多个智能算法的复合智能体模块；以及物理实体存储层，其与所述应用层通信连接，所述物理实体存储层包括通信连接的物理层存储实体主节点和物理层存储实体备用节点，所述物理层存储实体主节点与所述应用层智能主节点通信连接，其中，当所述物理层存储实体主节点发生故障时，所述物理层存储实体主节点和所述物理层存储实体备用节点能够通过状态倒换来使所述物理实体存储层正常工作。
  2. 根据权利要求 1 所述的焊接多形态数据智能体，其特征在于：所述具有多个智能算法的复合智能体包括多形态时序数据队列、图像特征表征智能算法体、异常检测智能算法体、时间校验一致性智能算法体和主备切片堆处理体。
  3. 根据权利要求 1 所述的焊接多形态数据智能体，其特征在于：所述物理实体存储层包括 I/O 读写功能和存储功能，所述 I/O 读写功能要求读写速度不低于主控制器下发指令的频率。
  4. 根据权利要求 1 所述的焊接多形态数据智能体，其特征在于：所述物理抗干扰隔离层为碳纳米管纤维材料，其用于吸收弧光的电磁辐射。
  5. 根据权利要求 1 所述的焊接多形态数据智能体，其特征在于：所述主控制器所采集的数据包括图像数据、焊机数据和机器本体运动数据，多个所述数据接口包括时序图像数据接口、主推时序数据接口、焊接时序数据接口和滑块时序数据接口，分别用于传输时序图像、主推数据、焊机数据和滑块数据。
  6. 一种高可靠抗弧光干扰的焊接多形态数据智能处理方法，其特征在于：所述焊接多形态数据智能处理方法包括如下步骤：
     
    构建主控制器、物理抗干扰隔离层、应用层、物理实体存储层以及通讯控制器；所述主控制器从爬行焊接机器人的多个传感设备实时采集数据；
     
    所述主控器通过多个数据接口将所采集的数据传输给应用层智能主节点的具有多个智能算法的复合智能体模块，其中，多个所述数据接口包括时序图像数据接口、主推时序数据接口、焊接时序数据接口和滑块时序数据接口，分别用于传输时序图像、主推数据、焊机数据和滑块数据；
     
    当所述复合智能体模块所接收到的时序图像、主推数据、焊机数据和滑块数据时，所述时序图像、主推数据、焊机数据和滑块数据进入处理队列；
     
    图像特征表征智能算法体通过图像特征表征算法将焊缝的图像数据进行转化处理，输出未焊接的焊缝参数，并输出已经焊接完成的焊接质量数据；
     
    异常检测智能算法体通过异常检测智能算法对转化处理后的数据进行异常甄别，并且对焊机数据和主推数据的原始数据做异常甄别和处理，并得到归一化形态数据；
     
    所述时间一致性校验算法体通过时间一致性校验算法将所述归一化形态数据进行逻辑校验，并得到时序单一形态数据；
     
    在状态开关处于开的状态下，把时序单一形态数据进行堆处理，形成可提取的时间轴回放数据，所述时间轴回放数据在切片堆栈队列中进行堆栈处理，并存储在物理实体存储层中；
     
    云端状态监听设备通过所述通讯控制器定时同步获取存储在所述物理实体存储层中的数据；
     
    其中，当应用层智能主节点发生故障时，所述应用层智能主节点和应用层智能备用节点进行状态倒换；并且当物理层存储实体主节点发生故障时，所述物理层存储实体主节点和物理层存储实体备用节点通过状态倒换来使所述物理实体存储层正常工作。
  7. 根据权利要求 6 所述的焊接多形态数据智能处理方法，其特征在于：还包括如下步骤：在进入云端同步的过程中，状态开关改为关闭状态，如果此时爬行焊接机器人处于运行态状态，时间一致性校验算法把逻辑校验后的归一化形态数据输送到切片堆栈备用队列；
    当云端同步结束之后，切片堆栈备用队列归并到切片堆栈队列，状态开关改为开启状态。
  8. 根据权利要求 6 所述的焊接多形态数据智能处理方法，其特征在于：通过图像特征表征算法将焊缝的图像数据进行转化处理包括如下步骤：先对 RGB 图像做二值化灰度处理，然后通过多个神经核去扫描，卷积计算出低维数据，通过多次低维度处理，最终输出若干 1*N 维数据，然后分别截取不同段表示各个数据，从而能够按照时间维度将焊缝图像提取出有效特征，并表征为结构化多维数据。
  9. 根据权利要求 6 所述的焊接多形态数据智能处理方法，其特征在于：对焊机数据和主推数据的原始数据做异常甄别和处理包括如下步骤：将焊机数据和主推数据的原始数据做异常时序数据的过滤，然后找出所有有问题的时间点集合，把该集合下所有数据全部做剔除处理，以得到归一化形态数据。
  10. 根据权利要求 6 所述的焊接多形态数据智能处理方法，其特征在于：所述主控制器所采集的数据包括图像数据、焊机数据和机器本体运动数据。
PCT/CN2020/137415 2020-12-18 2020-12-18 高可靠抗弧光干扰的焊接多形态数据智能体及处理方法 WO2022126564A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2020/137415 WO2022126564A1 (zh) 2020-12-18 2020-12-18 高可靠抗弧光干扰的焊接多形态数据智能体及处理方法

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2020/137415 WO2022126564A1 (zh) 2020-12-18 2020-12-18 高可靠抗弧光干扰的焊接多形态数据智能体及处理方法

Publications (1)

Publication Number Publication Date
WO2022126564A1 true WO2022126564A1 (zh) 2022-06-23

Family

ID=82059975

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/137415 WO2022126564A1 (zh) 2020-12-18 2020-12-18 高可靠抗弧光干扰的焊接多形态数据智能体及处理方法

Country Status (1)

Country Link
WO (1) WO2022126564A1 (zh)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN202067141U (zh) * 2011-05-27 2011-12-07 浙江师范大学 一种基于s3c2440的电焊机网络管理***
KR101184338B1 (ko) * 2011-03-18 2012-09-20 김상준 레이저 빔 강도의 선형 제어가 가능한 레이저 납땜 장치 및 레이저 용접 장치
CN104625334A (zh) * 2014-12-29 2015-05-20 黑龙江大学 粗管对焊自动控制装置及该控制装置的控制方法
CN111869165A (zh) * 2018-01-22 2020-10-30 西门子股份公司 用于控制和/或监控装置的方法和控制***
CN111940954A (zh) * 2020-08-14 2020-11-17 南京水木自动化科技有限公司 高可靠抗弧光干扰的焊接多形态数据智能体及处理方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR101184338B1 (ko) * 2011-03-18 2012-09-20 김상준 레이저 빔 강도의 선형 제어가 가능한 레이저 납땜 장치 및 레이저 용접 장치
CN202067141U (zh) * 2011-05-27 2011-12-07 浙江师范大学 一种基于s3c2440的电焊机网络管理***
CN104625334A (zh) * 2014-12-29 2015-05-20 黑龙江大学 粗管对焊自动控制装置及该控制装置的控制方法
CN111869165A (zh) * 2018-01-22 2020-10-30 西门子股份公司 用于控制和/或监控装置的方法和控制***
CN111940954A (zh) * 2020-08-14 2020-11-17 南京水木自动化科技有限公司 高可靠抗弧光干扰的焊接多形态数据智能体及处理方法

Similar Documents

Publication Publication Date Title
US10698777B2 (en) High availability scheduler for scheduling map-reduce searches based on a leader state
US20200394196A1 (en) Stream processing diagnostics
CN101067742B (zh) 量测方法以及虚拟量测***
CN108959400B (zh) 银行***历史数据清理方法及装置
CN106844161B (zh) 一种带状态流计算***中的异常监控及预测方法和***
CN112871703B (zh) 一种智能管理选煤平台及其方法
CN101136799B (zh) 一种实现通讯设备故障集中告警处理的方法
AU2019348202B2 (en) System and method for robotic agent management
CN106936812B (zh) 一种云环境下基于Petri网的文件隐私泄露检测方法
CN110532857A (zh) 基于多摄像头下的行为识别影像分析***
CN103324715A (zh) 一种灾备***可用性检测方法及装置
Shilpika et al. MELA: A visual analytics tool for studying multifidelity hpc system logs
US20090164443A1 (en) Database performance mining
WO2022126564A1 (zh) 高可靠抗弧光干扰的焊接多形态数据智能体及处理方法
CN111940954B (zh) 高可靠抗弧光干扰的焊接多形态数据智能处理方法
CN114153788A (zh) 一种执行器的可追溯控制方法、执行器和控制***
CN111045865A (zh) 一种基于块复制的实时同步方法及***
He et al. Graph based incident extraction and diagnosis in large-scale online systems
CN109885607A (zh) 一种工业海量非结构化数据处理方法及***
CN106250406A (zh) 一种日志处理方法
JP2022013579A (ja) 画像処理方法、装置、電子機器および記憶媒体
Quadri et al. Suspicious Activity Detection Using Convolution Neural Network
CN114677638B (zh) 一种基于深度学习和聚类人群异常聚集的检测方法
CN102221995A (zh) 地震数据处理作业的断点恢复方法
CN114741455A (zh) 一种中压不断电装置的降噪方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20965580

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20965580

Country of ref document: EP

Kind code of ref document: A1