立式注塑機要利用好機器大數(shù)據(jù)，需要突破幾類核心技術(shù)

首先是立式注塑機數(shù)據(jù)怎么管理的問題。海量的立式注塑機機器型數(shù)據(jù)如時間序列、時空數(shù)據(jù)等高速采集完成后，需要把它存下來，這涉及到數(shù)據(jù)有效打包、壓縮、放置的問題。數(shù)據(jù)存下來是為了被利用，這需要支持快速定位查詢到應(yīng)用需求的數(shù)據(jù)，而這又是一個如何建立高效的時空數(shù)據(jù)索引的問題。

數(shù)據(jù)存好管好了，下一個問題就是如何支持各種分析。做過實際分析的人都知道，分析絕不僅僅是開發(fā)一堆算法的問題。算法只是一小部分工作，大部分的工作是根據(jù)對業(yè)務(wù)問題的理解選取需要的數(shù)據(jù)，理解數(shù)據(jù)的特征，然后根據(jù)特征設(shè)計一個合適的模型和算法。這中間數(shù)據(jù)特征的理解對機器大數(shù)據(jù)來說是很難的。因為機器數(shù)據(jù)不能為人所直觀理解，需要交互特征工程。此外，從模型和算法的層面，機器數(shù)據(jù)往往是對一個物理系統(tǒng)的感知結(jié)果，而物理有許多機理性的原理存在，比如機械領(lǐng)域涉及力學(xué)原理，冶金領(lǐng)域涉及化學(xué)原理，因此立式注塑機的機器大數(shù)據(jù)的分析需要有機結(jié)合機理模型和數(shù)據(jù)統(tǒng)計模型。還有一個常常被忽略的問題是數(shù)據(jù)質(zhì)量的問題——如何把握數(shù)據(jù)質(zhì)量，如何修正數(shù)據(jù)質(zhì)量。

再談應(yīng)用的角度，如何更簡單地訪問數(shù)據(jù)和使用分析，特別是對領(lǐng)域?qū)＜?。在多源異質(zhì)數(shù)據(jù)，屏蔽數(shù)據(jù)集成關(guān)聯(lián)的問題，使得領(lǐng)域?qū)＜也恍枰畯?fù)雜的大數(shù)據(jù)技術(shù)和編程。

Vertical injection molding machine use good machine big data, need to break through several kinds of core technology

First of all, how to manage the data of vertical injection molding machine. The massive vertical injection molding machine data such as time series, spatio-temporal data and other high-speed acquisition after completion, it needs to save it, which involves the effective data packaging, compression, placement problems. Data is stored to be used, which requires the rapid positioning of data queries to the application requirements, and this is a problem of how to build efficient spatio-temporal data index.

The data is stored well, and the next question is how to support various analyses. Anyone who has actually done the analysis knows that analysis is not just a matter of developing algorithms. The algorithm is only a small part of the work, most of the work is based on the understanding of business problems, select the required data, understand the characteristics of the data, and then design a suitable model and algorithm according to the characteristics. This understanding of intermediate data features is difficult for machine big data. Because machine data can not be understood intuitively, interactive feature engineering is needed. In addition, from the model and the algorithm level, machine data is often on a physical world system and the physical world perception results, there is a lot of rational principle, such as machinery involved in the field of mechanical principle, and relates to the field of chemical metallurgy principle, so the analysis of machine data of vertical injection molding machine to combine model and mechanism statistical model. Another problem that is often overlooked is the problem of data quality - how to grasp the quality of data and how to correct the quality of data.

Talk about the application point of view, how to access data and use analysis more simply, especially for domain experts. In multi-source heterogeneous data, shielding Data Integration Association problem makes the domain experts do not need to understand complex big data technology and programming.