TWI700566B - Method for diagnosing abnormal dies in fastener making machine and computer program product thereof - Google Patents
Method for diagnosing abnormal dies in fastener making machine and computer program product thereof Download PDFInfo
- Publication number
- TWI700566B TWI700566B TW108143788A TW108143788A TWI700566B TW I700566 B TWI700566 B TW I700566B TW 108143788 A TW108143788 A TW 108143788A TW 108143788 A TW108143788 A TW 108143788A TW I700566 B TWI700566 B TW I700566B
- Authority
- TW
- Taiwan
- Prior art keywords
- pressure sensing
- sensing data
- mold
- sample
- abnormal
- Prior art date
Links
Images
Landscapes
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
- General Factory Administration (AREA)
Abstract
Description
本揭露是有關於一種模具異常診斷方法,且特別是有關於一種扣件成型機之模具異常診斷方法及其電腦程式產品。 The present disclosure relates to a method for diagnosing mold abnormalities, and in particular, to a method for diagnosing mold abnormalities of a fastener forming machine and a computer program product thereof.
由於目前扣件製程大多都是由人工抽檢方式檢驗加工過程及產品,故當產品出現瑕疵因子或模具異常時,機台並無法及時因應,導致浪費許多加工資源與成本。目前的扣件產業趨勢是朝向智能化線上監控發展,故若能有效掌握扣件製程模具失效過程,則可提高製程的良率及產能。 Since most of the fastener manufacturing process currently uses manual sampling to inspect the processing process and the product, when the product has a defect factor or a mold abnormality, the machine cannot respond in time, resulting in a waste of processing resources and costs. The current trend of the fastener industry is toward the development of intelligent online monitoring. Therefore, if the failure process of the fastener manufacturing process mold can be effectively grasped, the yield and productivity of the manufacturing process can be improved.
然而,目前扣件製程模具壽命的預測方式是透過長時間的監控扣件生產過程,並從中尋求模具失效過程,此種作法耗時且準確度不高。 However, the current method of predicting the life of the mold in the fastener manufacturing process is to monitor the fastener production process for a long time and seek the mold failure process from it, which is time-consuming and not accurate.
因此,本揭露之實施例之一目的是在提供一種模具異常診斷方法,其可在扣件成型機的加工過程中,快速判別模具的異常狀態,進而能夠提早採取適當措施進而避免不良加工的問題發生。 Therefore, one purpose of the embodiments of the present disclosure is to provide a mold abnormality diagnosis method, which can quickly determine the abnormal state of the mold during the processing of the fastener forming machine, so that appropriate measures can be taken early to avoid the problem of poor processing occur.
根據本揭露之實施例之上述目的,提出一種扣件成型機之模具異常診斷方法,包含以下步驟。安裝至少一壓力感測器至扣件成型機上。安裝正常樣本模具至扣件成型機上。使用扣件成型機和正常樣本模具分別處理第一樣本工件,而獲得一組正常樣本壓力感測資料,其中第一樣本工件對應至正常模具狀態。安裝異常樣本模具至扣件成型機上。使用扣件成型機和異常樣本模具分別處理第二樣本工件,而獲得一組異常樣本壓力感測資料,其中第二樣本工件對應至異常模具狀態。進行資料生成步驟,以利用生成對抗網路模型來生成複數組仿製壓力感測資料,其中生成對抗網路模型係使用所述之正常樣本壓力感測資料、所述之異常樣本壓力感測資料和複數個資料權重比並根據一生成對抗網路演算法來生成仿製壓力感測資料。仿製壓力感測資料係以一對一的方式對應至資料權重比,而資料權重比為正常樣本壓力感測資料對異常樣本壓力感測資料的成分比值,並代表仿製壓力感測資料所對應之複數個模具參考狀態。使用正常樣本壓力感測資料、異常樣本壓力感測資料、以及仿製壓力感測資料並根據自動編碼演算法建立編碼模型,其中編碼模型分別壓縮正常樣本壓力感測資料、異常樣本壓力感測資料、以及仿製壓力感測資料為複數組樣本編碼特徵,樣本編碼特徵對 應至正常模具狀態、異常模具狀態以及模具參考狀態。使用樣本編碼特徵和其對應之正常模具狀態、異常模具狀態以及模具參考狀態並根據推估演算法,來建立模具異常診斷模型。安裝標的模具至扣件成型機上。使用扣件成型機和標的模具處理標的工件,而獲得標的壓力感測資料。輸入標的壓力感測資料至編碼模型,以獲得標的編碼特徵。輸入標的編碼特徵至模具異常診斷模型中,而推估出針對標的模具之模具異常狀態。 According to the above objectives of the embodiments of the present disclosure, a method for diagnosing mold abnormalities of a fastener forming machine is provided, which includes the following steps. Install at least one pressure sensor on the fastener forming machine. Install the normal sample mold on the fastener forming machine. The fastener forming machine and the normal sample mold are used to process the first sample workpiece separately to obtain a set of normal sample pressure sensing data, wherein the first sample workpiece corresponds to the normal mold state. Install the abnormal sample mold on the fastener forming machine. The fastener forming machine and the abnormal sample mold are used to process the second sample workpiece separately to obtain a set of abnormal sample pressure sensing data, wherein the second sample workpiece corresponds to the abnormal mold state. A data generation step is performed to generate a complex array of imitation pressure sensing data using the generation confrontation network model, wherein the generation confrontation network model uses the normal sample pressure sensing data, the abnormal sample pressure sensing data and A plurality of data weight ratios are used to generate imitation pressure sensing data according to a generation confrontation network algorithm. The imitation pressure sensing data corresponds to the data weight ratio in a one-to-one manner, and the data weight ratio is the component ratio of the normal sample pressure sensing data to the abnormal sample pressure sensing data, and represents the corresponding value of the imitation pressure sensing data Multiple mold reference states. Use normal sample pressure sensing data, abnormal sample pressure sensing data, and imitation pressure sensing data to establish a coding model based on an automatic coding algorithm. The coding model compresses normal sample pressure sensing data, abnormal sample pressure sensing data, And the imitation pressure sensing data is a complex array of sample coding features, the sample coding feature pair Should be normal mold state, abnormal mold state and mold reference state. Use sample coding features and their corresponding normal mold state, abnormal mold state, and mold reference state to establish a mold abnormal diagnosis model based on the estimation algorithm. Install the target mold on the fastener forming machine. Use the fastener forming machine and the target mold to process the target workpiece and obtain the target pressure sensing data. Input the target pressure sensing data into the coding model to obtain the target coding feature. Input the coded features of the target into the mold abnormality diagnosis model, and estimate the abnormal state of the mold for the target mold.
根據本揭露之實施例之上述目的,提出另一種扣件成型機之模具異常診斷方法,包含以下步驟。獲取扣件成型機和正常樣本模具處理第一樣本工件時之一組正常樣本壓力感測資料,其中正常樣本壓力感測資料係由安裝在扣件成型機之上的至少一壓力感測器所獲得。獲取扣件成型機和異常樣本模具處理第二樣本工件時之一組異常樣本壓力感測資料,其中異常樣本壓力感測資料係由壓力感測器所獲得。進行資料生成步驟,以利用生成對抗網路模型來生成複數組仿製壓力感測資料,其中生成對抗網路模型係使用正常樣本壓力感測資料、異常樣本壓力感測資料和複數個資料權重比並根據一生成對抗網路演算法來生成仿製壓力感測資料,且仿製壓力感測資料係以一對一的方式對應至資料權重比,而資料權重比為正常樣本壓力感測資料對異常樣本壓力感測資料的成分比值,並代表仿製壓力感測資料所對應之複數個模具參考狀態。使用正常樣本壓力感測資料、異常樣本壓力感測資料、以及仿製壓力感測資料並根據自動編碼演算 法建立編碼模型,其中編碼模型分別壓縮正常樣本壓力感測資料、異常樣本壓力感測資料、以及仿製壓力感測資料為複數組樣本編碼特徵,樣本編碼特徵對應至正常模具狀態、異常模具狀態以及模具參考狀態。使用樣本編碼特徵和其對應之正常模具狀態、異常模具狀態以及模具參考狀態並根據推估演算法,來建立模具異常診斷模型。使用扣件成型機和標的模具處理標的工件,而獲得標的壓力感測資料。輸入標的壓力感測資料至編碼模型,以獲得標的編碼特徵。輸入標的編碼特徵至模具異常診斷模型中,而推估出針對標的模具之模具異常狀態。 According to the above objectives of the embodiments of the present disclosure, another method for diagnosing mold abnormalities of a fastener forming machine is proposed, which includes the following steps. Obtain a set of normal sample pressure sensing data when the fastener forming machine and the normal sample mold are processing the first sample workpiece, wherein the normal sample pressure sensing data is provided by at least one pressure sensor installed on the fastener forming machine Obtained. Obtain a set of abnormal sample pressure sensing data when the fastener forming machine and the abnormal sample mold process the second sample workpiece, wherein the abnormal sample pressure sensing data is obtained by the pressure sensor. Perform a data generation step to generate a complex array of imitation pressure sensing data using the generative confrontation network model, where the generation of the confrontation network model uses normal sample pressure sensing data, abnormal sample pressure sensing data, and a combination of multiple data weight ratios The imitation pressure sensing data is generated according to a generative confrontation network algorithm, and the imitation pressure sensing data is corresponding to the data weight ratio in a one-to-one manner, and the data weight ratio is the normal sample pressure sensing data versus the abnormal sample pressure sensing The component ratio of the measured data and represents the reference state of multiple molds corresponding to the imitation pressure sensing data. Use normal sample pressure sensing data, abnormal sample pressure sensing data, and imitation pressure sensing data and calculate based on automatic coding The coding model is established by the method, in which the coding model compresses the normal sample pressure sensing data, the abnormal sample pressure sensing data, and the imitation pressure sensing data as a complex array of sample coding features. The sample coding features correspond to the normal mold state, abnormal mold state, and Mold reference status. Use sample coding features and their corresponding normal mold state, abnormal mold state, and mold reference state to establish a mold abnormal diagnosis model based on the estimation algorithm. Use the fastener forming machine and the target mold to process the target workpiece and obtain the target pressure sensing data. Input the target pressure sensing data into the coding model to obtain the target coding feature. Input the coded features of the target into the mold abnormality diagnosis model, and estimate the abnormal state of the mold for the target mold.
依據本揭露之一實施例,上述之每一組正常樣本壓力感測資料及異常樣本壓力感測資料是由壓力感測器所測得之實際壓力對時間的關係曲線。每一仿製壓力感測資料為生成對抗網路模型所仿製之仿製壓力對時間的關係曲線,且仿製壓力對時間的關係曲線的分布位置介於對應正常樣本壓力感測資料之實際壓力對時間的關係曲線與對應異常樣本壓力感測資料之實際壓力對時間的關係曲線之間。 According to an embodiment of the present disclosure, each set of normal sample pressure sensing data and abnormal sample pressure sensing data is a curve of actual pressure versus time measured by a pressure sensor. Each simulated pressure sensing data is to generate a simulated pressure versus time curve that is simulated by a network model, and the distribution position of the simulated pressure versus time curve is between the actual pressure versus time corresponding to the normal sample pressure sensing data Between the relationship curve and the actual pressure versus time relationship curve corresponding to the abnormal sample pressure sensing data.
依據本揭露之一實施例,推估演算法包括隨機森林(Random Forest)演算法、支持向量機(Support Vector Machines,SVM)演算法或深度神經網路(Deep neural network,DNN)演算法。 According to an embodiment of the disclosure, the estimation algorithm includes a random forest (Random Forest) algorithm, a support vector machine (SVM) algorithm or a deep neural network (DNN) algorithm.
依據本揭露之一實施例,其中編碼模型包含壓縮器以及解碼器。其中,壓縮器可將每一組正常樣本壓力感測資料、異常樣本壓力感測資料、仿製壓力感測資料、以及 標的壓力感測資料壓縮後形成壓縮資料。解碼器可將壓縮資料解碼還原成對應每一個壓縮資料之解碼資料。 According to an embodiment of the disclosure, the encoding model includes a compressor and a decoder. Among them, the compressor can combine each set of normal sample pressure sensing data, abnormal sample pressure sensing data, imitation pressure sensing data, and The target pressure sensing data is compressed to form compressed data. The decoder can decode and restore compressed data into decoded data corresponding to each compressed data.
根據本揭露之實施例之上述目的,另提出一種用於診斷扣件成型機之模具異常狀態之電腦程式產品。當電腦載入此電腦程式產品並執行後,可完成如前述之扣件成型機之模具異常診斷方法。 According to the above-mentioned purpose of the embodiments of the present disclosure, another computer program product for diagnosing the abnormal state of the mold of the fastener forming machine is provided. When the computer loads this computer program product and executes it, it can complete the mold abnormal diagnosis method of the fastener forming machine mentioned above.
由上述可知,本揭露是以正常樣本壓力感測資料與異常樣本壓力感測資料為基礎、並透過混合不同比例的正常樣本與異常樣本之成分比值,使用生成對抗網路模型來生成對應正常模具至異常模具之漸變資料的仿製壓力感測資料,再透過這些資料來建立模具異常診斷模型,藉以達到預測標的模具的異常狀態。另一方面,本揭露係利用自動編碼演算法來建立編碼模型,並利用編碼模型尋找出能夠代表扣件成型機所取得的壓力感測資料之關鍵特徵,然後再利用這些關鍵特徵與其對應模具狀態建立模具異常預測模型,進而減少大量資料處理時間。 It can be seen from the above that this disclosure is based on the normal sample pressure sensing data and the abnormal sample pressure sensing data, and by mixing the component ratios of the normal sample and the abnormal sample in different proportions, the generation confrontation network model is used to generate the corresponding normal mold The imitation pressure sensing data of the gradual data of the abnormal mold is used to establish a mold abnormal diagnosis model through these data, so as to predict the abnormal state of the target mold. On the other hand, this disclosure uses an automatic coding algorithm to establish a coding model, and uses the coding model to find key features that can represent the pressure sensing data obtained by the fastener forming machine, and then uses these key features and their corresponding mold states Establish a mold abnormality prediction model, thereby reducing a lot of data processing time.
100‧‧‧模具異常診斷方法 100‧‧‧Mold abnormal diagnosis method
101~112‧‧‧步驟 101~112‧‧‧Step
200‧‧‧扣件成型機 200‧‧‧Fastener Forming Machine
210‧‧‧壓力感測器 210‧‧‧Pressure Sensor
220‧‧‧第一模座 220‧‧‧First mold base
220a‧‧‧第一模具 220a‧‧‧First mold
222‧‧‧第二模座 222‧‧‧Second mold base
222a‧‧‧第二模具 222a‧‧‧Second mold
300‧‧‧編碼模型 300‧‧‧Coding Model
310‧‧‧壓縮器 310‧‧‧Compressor
320‧‧‧解碼器 320‧‧‧Decoder
500‧‧‧模具異常診斷方法 500‧‧‧Mold abnormal diagnosis method
501~508‧‧‧步驟 501~508‧‧‧Step
A‧‧‧連接點 A‧‧‧Connecting point
為了更完整了解實施例及其優點,現參照結合所附圖式所做之下列描述,其中: For a more complete understanding of the embodiments and their advantages, reference is now made to the following description in conjunction with the accompanying drawings, in which:
〔圖1A〕及〔圖1B〕為本揭露之一實施方式之一種扣件成型機之模具異常診斷方法的流程示意圖; [FIG. 1A] and [FIG. 1B] are schematic flow diagrams of a method for diagnosing mold abnormalities of a fastener forming machine according to an embodiment of the disclosure;
〔圖2〕係繪示依照本揭露之一實施方式之一種扣件成型機之局部裝置示意圖; [Figure 2] is a schematic diagram of a partial device of a fastener forming machine according to an embodiment of the present disclosure;
〔圖3A〕至〔圖3D〕為生成對抗網路模型根據不同之資料權重比所生成之仿製壓力對時間的關係曲線圖; [Figure 3A] to [Figure 3D] are graphs of the relationship between imitation pressure and time generated by the generation of confrontation network models according to different data weight ratios;
〔圖4〕係繪示依照本揭露之一實施方式之編碼模型的運作示意圖; [Figure 4] is a schematic diagram showing the operation of the coding model according to an embodiment of the present disclosure;
〔圖5〕係繪示依照本揭露之一實施方式之編碼模型與預測模型的運作示意圖;以及 [Fig. 5] is a schematic diagram showing the operation of the coding model and the prediction model according to an embodiment of the present disclosure; and
〔圖6〕係繪示依照本揭露之一實施方式之另一種扣件成型機之模具異常診斷方法的流程示意圖。 [FIG. 6] is a schematic flow diagram of another method for diagnosing mold abnormalities of a fastener forming machine according to an embodiment of the present disclosure.
請同時參照圖1A、圖1B及圖2,其中圖1A及圖1B為本揭露之一實施方式之一種扣件成型機之模具異常診斷方法的流程示意圖,圖2係繪示依照本揭露之一實施方式之一種扣件成型機之局部裝置示意圖。如圖2所示,本實施方式之扣件成型機200主要包含第一模座220以及第二模座222。第一模座220上裝設有第一模具220a,第二模座222上裝設有第二模具222a。在一示範例子中,第一模座220可為公模座,第二模座222可為母模座,且第一模座220連接沖壓裝置,沖壓裝置可帶動第一模座220向第二模座222移動,並使第一模具220a沖壓固定在第二模具222a上的胚料,以形成扣件成品或半成品。
Please refer to FIGS. 1A, 1B, and 2 at the same time. Among them, FIGS. 1A and 1B are a schematic flowchart of a method for diagnosing a mold abnormality of a fastener forming machine according to an embodiment of the present disclosure. A schematic diagram of a partial device of a fastener forming machine of the embodiment. As shown in FIG. 2, the
如圖1A、圖1B及圖2所示,本揭露實施例之模
具異常診斷方法100的流程分圖1A及圖1B繪製,其中圖1A之連接點A與圖1B中之連接點A對應。本實施方式之模具異常診斷方法100主要包含以下步驟。首先,進行步驟101,以將至少一壓力感測器210安裝至扣件成型機200的第一模座220上。接著,進行步驟102,以將正常樣本模具作為第一模具220a安裝至扣件成型機上的第一模座220上。然後,進行步驟103,以使用扣件成型機200和正常樣本模具分別處理複數個第一樣本工件,而獲得一組正常樣本壓力感測資料。其中,第一樣本工件對應至正常模具狀態。在一些實施例中,正常樣本壓力感測資料為鍛造力波形變化圖。在本實施例中,正常樣本壓力感測資料是當正常樣本模具在進行扣件成形步驟時,由壓力感測器210所實際測得之壓力對時間的關係曲線圖。
As shown in FIG. 1A, FIG. 1B and FIG. 2, the model of the embodiment of the present disclosure
The flow of the
接著,進行步驟104,以將異常樣本模具作為第一模具220a安裝至扣件成型機200上的第一模座220上。然後,進行步驟105,以使用扣件成型機200和異常樣本模具分別處理複數個第二樣本工件,而獲得一組異常樣本壓力感測資料。其中,第二樣本工件對應至異常模具狀態。在本實施例中,異常模具狀態可包含例如公模缺角狀態、母模異物狀態、胚料異常狀態、與潤滑不足狀態。在一些實施例中,異常樣本壓力感測資料同樣為鍛造力波形變化圖。在本實施例中,異常樣本壓力感測資料是當正異常樣本模具在進行扣件成形步驟時,由壓力感測器210所實際測得之壓力對時間的關係曲線圖。欲陳明者,本實施例主要是以第一模
座220上的第一模具220a作為測量標的,故可直接設定第二模座222上的第二模具222a為正常模具。在其他實施例中,若以第二模座222上的第二模具222a作為量測標的的話,亦可將壓力感測器210安裝在第二模座222上,並設定第一模座220上的第一模具220a為正常模具。
Next, proceed to step 104 to install the abnormal sample mold as the
在獲得正常樣本壓力感測資料與異常樣本壓力感測資料後,接著進行步驟106,以進行資料生成步驟。在資料生成步驟中,可利用生成對抗網路(Generative Adversarial Network,GAN)模型來生成複數組仿製壓力感測資料。在本實施例中,仿製壓力感測資料為生成對抗網路模型所仿製之壓力對時間的關係曲線。其中,生成對抗網路模型係使用正常樣本壓力感測資料、異常樣本壓力感測資料和複數個資料權重比並根據一生成對抗網路演算法來生成仿製壓力感測資料。在一些實施例中,資料權重比可代表仿製壓力感測資料所對應之複數個模具參考狀態。在本實施例中,資料權重比是由正常模具與異常模具的資料所定義。舉例而言,對於已知之正常模具而言,正常成分比異常成分的比值為1,故可將資料權重比為1對應至一正常模具狀態。對於已知異常模具而言,正常成分比異常成分的比值為0,故可將資料權重比為0對應至一異常模具狀態。因此,資料權重比在0~1之間的數值則可對應至數個模具參考狀態,且這些模具參考狀態可代表模具異常過程。也就是說,仿製壓力感測資料係以一對一的方式對應至資料權重比,而這些資料權重比為正常樣本壓力感測資料對異常樣本壓力
感測資料的成分比值,並代表仿製壓力感測資料所對應之模具參考狀態。值得一提的是,生成對抗網路演算法的原理為本領域中的技術人員所熟知,故於此不再贅述。
After obtaining the normal sample pressure sensing data and the abnormal sample pressure sensing data,
請參照下表一,在本實施例中,下表一是利用不同正常成分與異常成分的比值來作為資料權重比,以作為模具異常過程的參考狀態。利用生成對抗網路模型中的生成器根據資料權重比來生成仿製資料,再透過生成對抗網路模型中的判別器來分辨資料的真假,並給出一個回饋,且生成器可根據判別器的回饋來訓練並調整模型參數,反覆此過程直到生成器能夠生成接近真實資料的仿製壓力感測資料為止。 Please refer to Table 1 below. In this embodiment, Table 1 uses the ratios of different normal components to abnormal components as the weight ratio of the data and serves as the reference state of the abnormal process of the mold. Use the generator in the generative confrontation network model to generate imitation data according to the weight ratio of the data, and then use the discriminator in the generated confrontation network model to distinguish the true and false of the data, and give a feedback, and the generator can be based on the discriminator To train and adjust the model parameters, repeat the process until the generator can generate imitation pressure sensing data close to the real data.
另請參照圖3A至圖3D所示,圖3A至圖3D為生成對抗網路模型根據不同之資料權重比所生成之仿製壓力對時間的關係曲線圖。在圖3A及圖3D中,虛線為使用正常樣本模具處理第一樣本工件時,壓力感測器所測得之實際壓 力對時間之正常樣本壓力感測資料;粗實線為使用異常樣本模具處理第二樣本工件時,壓力感測器所測得之實際壓力對時間之異常樣本壓力感測資料;位於虛線與粗實線之間的細實線則是生成對抗網路模型所仿製之仿製壓力對時間的關係曲線。由圖3A至圖3D可知,生成對抗網路模型所生成之仿製壓力對時間的關係曲線的分布位置是介於正常樣本壓力感測資料之壓力對時間的關係曲線與異常樣本壓力感測資料之壓力對時間的關係曲線之間。此外,從圖3A至圖3D之仿製壓力感測資料可知,當資料權重比越高時,代表模具正常成分較高,故所仿製的壓力對時間的關係曲線越接近正常樣本壓力感測資料(例如圖3A所示),相反地,當資料權重比越低時,代表模具正常成分較低,故所仿製的壓力對時間的關係曲線越接近異常樣本壓力感測資料(例如圖3D所示)。由此可知,利用生成對抗網路模型可產生具漸變特性的仿製壓力感測資料。 Please also refer to FIG. 3A to FIG. 3D. FIG. 3A to FIG. 3D are graphs showing the relationship between imitation pressure and time generated by the generating confrontation network model according to different data weight ratios. In Figure 3A and Figure 3D, the dotted line is the actual pressure measured by the pressure sensor when the normal sample mold is used to process the first sample workpiece. Normal sample pressure sensing data of force versus time; the thick solid line is the abnormal sample pressure sensing data of actual pressure versus time measured by the pressure sensor when the abnormal sample mold is used to process the second sample workpiece; The thin solid line between the solid lines is the relationship curve of the imitation pressure against time that is imitated by the generated confrontation network model. From Figures 3A to 3D, it can be seen that the distribution position of the imitation pressure versus time curve generated by the anti-network model is between the pressure versus time curve of the normal sample pressure sensing data and the abnormal sample pressure sensing data The relationship between pressure and time. In addition, from the simulated pressure sensing data in Figures 3A to 3D, it can be seen that the higher the weight ratio of the data, the higher the normal component of the mold, so the simulated pressure versus time curve is closer to the normal sample pressure sensing data ( For example, as shown in Figure 3A), on the contrary, when the data weight ratio is lower, it means that the normal component of the mold is lower, so the simulated pressure versus time curve is closer to the abnormal sample pressure sensing data (for example, as shown in Figure 3D) . It can be seen that using the generative confrontation network model can generate imitation pressure sensing data with gradual characteristics.
請同時參照圖1A、圖1B及圖4,其中圖4係繪示依照本揭露之一實施方式之編碼模型的運作示意圖。在獲得對應至正常模具狀態之正常樣本壓力感測資料、對應至異常模具狀態之異常樣本壓力感測資料、以及對應至可代表模具異常過程之數個模具參考狀態的仿製壓力感測資料後,可進行步驟107,使用正常樣本壓力感測資料、異常樣本壓力感測資料、以及仿製壓力感測資料並根據自動編碼(AutoEncoder)演算法建立編碼模型300。AutoEncoder演算法主要是使用對稱的模型結構,將原始資料進行壓縮和
解壓縮資料訓練模型,當解壓縮後的資料趨近於原始資料,則在壓縮後所產生的資料可直接作為原始資料的代表特徵。在本實施例中,編碼模型300可分別壓縮正常樣本壓力感測資料、異常樣本壓力感測資料、以及仿製壓力感測資料為複數個樣本編碼特徵,這些樣本編碼特徵一對一對應至正常模具狀態、異常模具狀態以及模具參考狀態。AutoEncoder演算法的原理為本領域中的技術人員所熟知,故於此不再贅述。
Please refer to FIG. 1A, FIG. 1B and FIG. 4 at the same time. FIG. 4 is a schematic diagram of the operation of the coding model according to an embodiment of the present disclosure. After obtaining the normal sample pressure sensing data corresponding to the normal mold state, the abnormal sample pressure sensing data corresponding to the abnormal mold state, and the imitation pressure sensing data corresponding to several mold reference states that can represent the abnormal process of the mold,
請繼續參照圖1A、圖1B及圖4,編碼模型300包含壓縮器310以及解碼器320。其中,壓縮器310可將每一個正常樣本壓力感測資料、異常樣本壓力感測資料、以及仿製壓力感測資料壓縮後形成壓縮資料。解碼器320可將壓縮資料解碼還原成對應每一個壓縮資料之解碼資料。如圖4所示,若要判斷壓縮資料是否可作為代表樣本壓力感測資料的樣本編碼特徵,可透過比對解碼資料與其對應之樣本壓力感測資料的差異,若差異小於門檻值時,代表壓縮資料可做為對應之樣本壓力感測資料之樣本編碼特徵。若差異大於門檻值時,則調整編碼模型的相關參數。
Please continue to refer to FIGS. 1A, 1B and 4, the
另請一併參照圖1A、圖1B及圖5,其中圖5係繪示依照本揭露之一實施方式之編碼模型與預測模型的運作示意圖。在取得樣本編碼特徵及其對應的模具狀態後,可進行步驟108,以使用樣本編碼特徵和其對應之正常模具狀態、異常模具狀態以及模具參考狀態並根據推估演算法,來建立模具異常診斷模型400。在一些例子中,推估演算法包
括隨機森林(Random Forest,RF)演算法、支持向量機(Support Vector Machines,SVM)演算法或深度神經網路(Deep neural network,DNN)演算法。其中,RF演算法、SVM演算法與DNN演算法的原理為本領域中的技術人員所熟知,故於此不再贅述。在一些例子中,透過RF演算法來建立模具異常預測模型,可用於快速診斷多種模具異常模式。
Please also refer to FIG. 1A, FIG. 1B, and FIG. 5. FIG. 5 is a schematic diagram of the operation of the coding model and the prediction model according to an embodiment of the present disclosure. After obtaining the sample code feature and its corresponding mold state, step 108 can be performed to use the sample code feature and its corresponding normal mold state, abnormal mold state, and mold reference state to establish a mold abnormality diagnosis based on the
請繼續參照圖1A、圖1B及圖5,在建立完模具異常診斷模型400後,可進行步驟109,將標的模具作為第一模具220a安裝至如圖2所示之扣件成型機200的第一模座220上。在本實施例中,標的模具為未知狀態之待預測的模具。接著,進行步驟110,以使用扣件成型機200和標的模具處理標的工件,而獲得標的壓力感測資料。在本實施例中,標的壓力感測資料是當標的模具在進行扣件成形步驟時,由壓力感測器210所測得之壓力對時間的關係曲線圖。
Please continue to refer to FIGS. 1A, 1B, and 5. After the mold
請繼續參照圖1A、圖1B及圖5,在獲得標的壓力感測資料後,接著進行步驟111,輸入標的壓力感測資料至編碼模型300,以獲得標的編碼特徵,此標的編碼特徵可作為標的壓力感測資料的代表特徵。在獲得標的壓力感測資料後,可進行步驟112。在步驟110中,輸入標的編碼特徵至模具異常診斷模型400中,而推估出針對標的模具所對應之模具異常狀態。
Please continue to refer to Figures 1A, 1B and 5, after obtaining the target pressure sensing data, proceed to step 111 to input the target pressure sensing data to the
請再次參照圖1A、圖1B及圖5,本揭露之模具異常診斷方法100主要分為訓練階段以及推估階段。在訓練
階段中,是以正常樣本壓力感測資料、異常樣本壓力感測資料、以及仿製壓力感測資料作為圖5所示之壓力感測資料輸入至編碼模型300中,而透過編碼模型300將正常樣本壓力感測資料、異常樣本壓力感測資料、以及仿製壓力感測資料為樣本編碼特徵後,可將樣本編碼特徵連同對應之正常模具狀態、異常模具狀態以及模具參考狀態,共同建立模具異常診斷模型400。在模具異常診斷模型400建立後,則可進入推估階段。在推估階段中,是以標的壓力感測資料作為圖5所示之壓力感測資料輸入至編碼模型300中,當一個未知狀態的標的模型在使用時,編碼模型300同樣可將壓力感測器所取得的標的壓力感測資料壓縮為標的編碼特徵。使用標的編碼特徵作為模具異常診斷模型400的輸入值後,模具異常診斷模型400則可推估出標的模型的模具異常診斷模型400。
Please refer to FIG. 1A, FIG. 1B and FIG. 5 again. The mold
另請同時參照圖2及圖6,其中圖6係繪示依照本揭露之一實施方式之另一種扣件成型機之模具異常診斷方法的流程示意圖。本實施方式之模具異常診斷方法500主要包含以下步驟。首先,進行步驟501,獲取扣件成型機200和正常樣本模具處理第一樣本工件時之一組正常樣本壓力感測資料。其中,正常樣本壓力感測資料係由安裝在如圖2所示之扣件成型機200之第一模座220上的壓力感測器210所獲得。接著,進行步驟502,獲取扣件成型機和異常樣本模具處理第二樣本工件時之一組異常樣本壓力感測資料。其中,異常樣本壓力感測資料係由安裝在如圖2所示之扣件成
型機之第一模座220上的壓力感測器210所獲得。
Please also refer to FIGS. 2 and 6 at the same time. FIG. 6 is a schematic flowchart of another method for diagnosing mold abnormalities of a fastener forming machine according to an embodiment of the present disclosure. The mold
然後,進行步驟503,進行資料生成步驟,以利用生成對抗網路模型來生成複數組仿製壓力感測資料。其中,生成對抗網路模型係使用正常樣本壓力感測資料、異常樣本壓力感測資料和複數個資料權重比來生成仿製壓力感測資料。仿製壓力感測資料係以一對一的方式對應至資料權重比,而資料權重比代表仿製壓力感測資料所對應之數個模具參考狀態。接著,進行步驟504,使用正常樣本壓力感測資料、異常樣本壓力感測資料、以及仿製壓力感測資料並根據自動編碼演算法建立編碼模型,其中編碼模型分別壓縮正常樣本壓力感測資料、異常樣本壓力感測資料、以及仿製壓力感測資料為複數組樣本編碼特徵,樣本編碼特徵對應至正常模具狀態、異常模具狀態以及模具參考狀態。 Then, proceed to step 503 to perform a data generation step to generate a complex array of imitation pressure sensing data using the generative confrontation network model. Among them, the generating confrontation network model uses normal sample pressure sensing data, abnormal sample pressure sensing data, and multiple data weight ratios to generate imitation pressure sensing data. The imitation pressure sensing data corresponds to the data weight ratio in a one-to-one manner, and the data weight ratio represents a number of mold reference states corresponding to the imitation pressure sensing data. Next, proceed to step 504 to use the normal sample pressure sensing data, the abnormal sample pressure sensing data, and the imitation pressure sensing data to establish a coding model according to the automatic coding algorithm. The coding model compresses the normal sample pressure sensing data and the abnormal pressure sensing data respectively. The sample pressure sensing data and the imitation pressure sensing data are a complex array of sample coding features, and the sample coding features correspond to the normal mold state, the abnormal mold state, and the mold reference state.
然後,進行步驟505,以使用樣本編碼特徵和其對應之正常模具狀態、異常模具狀態以及模具參考狀態並根據推估演算法,來建立如圖5所示之模具異常診斷模型400。在步驟505後,接著進行步驟506,以使用扣件成型機和標的模具分別處理數個標的工件,而獲得標的壓力感測資料。在步驟506後,接著進行步驟507,輸入標的壓力感測資料至如圖4及圖5所示之編碼模型300,以獲得標的編碼特徵。在獲得標的壓力感測資料後,可進行步驟508。在步驟508中,輸入標的編碼特徵至模具異常診斷模型400中,而推估出針對標的模具所對應之模具異常狀態。
Then, step 505 is performed to use the sample code feature and its corresponding normal mold state, abnormal mold state, and mold reference state to establish a mold
欲陳明者,圖6所示實施方式之步驟501、502、
501、502、503、504、505、506、507及508的具體進行方式分別與圖1A、圖1B所示的步驟103、105、106、107、108、110、111及112相同,故於此不再贅述。
For those who want to explain,
可理解的是,本揭露之模具異常診斷方法500為以上所述之實施步驟。上述實施例所說明的各實施步驟的次序可依實際需要而調動、結合或省略。上述實施例可利用電腦程式產品來實現,其可包含儲存有多個指令之機器可讀取媒體,這些指令可程式化(programming)電腦來進行上述實施例中的步驟。機器可讀取媒體可為但不限定於軟碟、光碟、唯讀光碟、磁光碟、唯讀記憶體、隨機存取記憶體、可抹除可程式唯讀記憶體(EPROM)、電子可抹除可程式唯讀記憶體(EEPROM)、光卡(optical card)或磁卡、快閃記憶體、或任何適於儲存電子指令的機器可讀取媒體。再者,本揭露之實施例也可做為電腦程式產品來下載,其可藉由使用通訊連接(例如網路連線之類的連接)之資料訊號來從遠端電腦轉移本揭露之電腦程式產品至請求電腦。
It is understandable that the mold
由上述實施方式可知,本揭露是以正常樣本壓力感測資料與異常樣本壓力感測資料為基礎、並透過混合不同比例的正常樣本與異常樣本之成分比值,使用生成對抗網路模型來生成對應正常模具至異常模具之漸變資料的仿製壓力感測資料,再透過這些資料來建立模具異常診斷模型,藉以達到預測標的模具的異常狀態。 It can be seen from the above-mentioned embodiments that the present disclosure is based on the normal sample pressure sensing data and the abnormal sample pressure sensing data, and by mixing the component ratios of the normal sample and the abnormal sample in different proportions, the generation confrontation network model is used to generate the corresponding The imitation pressure sensing data of the gradual data from the normal mold to the abnormal mold is used to establish a mold abnormal diagnosis model through these data, so as to predict the abnormal state of the target mold.
另一方面,本揭露係利用自動編碼演算法來建立編碼模型,並利用編碼模型尋找出能夠代表扣件成型機所 取得的壓力感測資料之關鍵特徵,然後再利用這些關鍵特徵與其對應模具狀態建立模具異常預測模型,進而減少大量資料處理時間。 On the other hand, this disclosure uses an automatic coding algorithm to establish a coding model, and uses the coding model to find a representative fastener forming machine location. The key features of the obtained pressure sensing data are then used to build a mold abnormal prediction model using these key features and their corresponding mold states, thereby reducing a lot of data processing time.
雖然本揭露之實施例已以實施例揭露如上,然其並非用以限定本揭露之實施例,任何所屬技術領域中具有通常知識者,在不脫離本揭露之實施例的精神和範圍內,當可作些許的更動與潤飾,故本揭露之實施例的保護範圍當視後附的申請專利範圍所界定者為準。 Although the embodiments of the present disclosure have been disclosed as above, they are not intended to limit the embodiments of the present disclosure. Anyone with ordinary knowledge in the relevant technical field should not depart from the spirit and scope of the embodiments of the present disclosure. Some changes and modifications can be made, so the protection scope of the embodiments of this disclosure shall be subject to the scope of the attached patent application.
100‧‧‧模具異常診斷方法 100‧‧‧Mold abnormal diagnosis method
101~107‧‧‧步驟 101~107‧‧‧Step
A‧‧‧連接點 A‧‧‧Connecting point
Claims (9)
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
TW108143788A TWI700566B (en) | 2019-11-29 | 2019-11-29 | Method for diagnosing abnormal dies in fastener making machine and computer program product thereof |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
TW108143788A TWI700566B (en) | 2019-11-29 | 2019-11-29 | Method for diagnosing abnormal dies in fastener making machine and computer program product thereof |
Publications (2)
Publication Number | Publication Date |
---|---|
TWI700566B true TWI700566B (en) | 2020-08-01 |
TW202121087A TW202121087A (en) | 2021-06-01 |
Family
ID=73003392
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
TW108143788A TWI700566B (en) | 2019-11-29 | 2019-11-29 | Method for diagnosing abnormal dies in fastener making machine and computer program product thereof |
Country Status (1)
Country | Link |
---|---|
TW (1) | TWI700566B (en) |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2005010522A2 (en) * | 2003-07-18 | 2005-02-03 | Rosemount Inc. | Process diagnostics |
TW200805018A (en) * | 2005-07-11 | 2008-01-16 | Brooks Automation Inc | Intelligent condition-monitoring and fault diagnostic system for robotized manufacturing tools |
TW201035553A (en) * | 2008-11-12 | 2010-10-01 | Qiagen Gaithersburg Inc | Automated assay and system |
TW201439928A (en) * | 2013-04-03 | 2014-10-16 | Lumax Internat Corp Ltd | Method and system for equipment maintenance |
US20170160125A1 (en) * | 2015-12-04 | 2017-06-08 | Seoul National University R&Db Foundation | Apparatus and method for diagnosing rotor shaft |
CN207311477U (en) * | 2017-08-31 | 2018-05-04 | 广州港集团有限公司 | The detection system of positioning and fastener defects detection is identified for sleeper |
TW201820064A (en) * | 2016-11-29 | 2018-06-01 | 財團法人工業技術研究院 | Prediction model building method and associated predicting method and computer software product |
CN108334033A (en) * | 2018-02-28 | 2018-07-27 | 中国科学院重庆绿色智能技术研究院 | Punching machine group failure prediction method and its system based on Internet of Things and machine learning |
-
2019
- 2019-11-29 TW TW108143788A patent/TWI700566B/en active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2005010522A2 (en) * | 2003-07-18 | 2005-02-03 | Rosemount Inc. | Process diagnostics |
TW200805018A (en) * | 2005-07-11 | 2008-01-16 | Brooks Automation Inc | Intelligent condition-monitoring and fault diagnostic system for robotized manufacturing tools |
TW201035553A (en) * | 2008-11-12 | 2010-10-01 | Qiagen Gaithersburg Inc | Automated assay and system |
TW201439928A (en) * | 2013-04-03 | 2014-10-16 | Lumax Internat Corp Ltd | Method and system for equipment maintenance |
US20170160125A1 (en) * | 2015-12-04 | 2017-06-08 | Seoul National University R&Db Foundation | Apparatus and method for diagnosing rotor shaft |
TW201820064A (en) * | 2016-11-29 | 2018-06-01 | 財團法人工業技術研究院 | Prediction model building method and associated predicting method and computer software product |
CN207311477U (en) * | 2017-08-31 | 2018-05-04 | 广州港集团有限公司 | The detection system of positioning and fastener defects detection is identified for sleeper |
CN108334033A (en) * | 2018-02-28 | 2018-07-27 | 中国科学院重庆绿色智能技术研究院 | Punching machine group failure prediction method and its system based on Internet of Things and machine learning |
Also Published As
Publication number | Publication date |
---|---|
TW202121087A (en) | 2021-06-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP4573036B2 (en) | Inspection apparatus and inspection method | |
CN109308063B (en) | Fermentation process stage division method based on score matrix | |
EP3623097A1 (en) | Soldering process parameter suggestion method and system thereof | |
CN108445838B (en) | Numerical control machine tool machining quality analysis method, classifier and equipment | |
TW200951752A (en) | A method for prognostic maintenance in semiconductor manufacturing | |
Cudney et al. | Applying the Mahalanobis–Taguchi system to vehicle handling | |
CN102601881B (en) | Method for monitoring on-line quality and updating prediction model of rubber hardness | |
CN114444231B (en) | Online self-adaptive prediction method, device, equipment and medium for residual life of mold | |
Rose | Estimation of the cycle time distribution of a wafer fab by a simple simulation model | |
CN106656669A (en) | Equipment parameter abnormity detection system and method based on self-adaptive setting of threshold | |
TWI700566B (en) | Method for diagnosing abnormal dies in fastener making machine and computer program product thereof | |
TWI709922B (en) | A model-based machine learning system | |
US10432800B1 (en) | Apparatus and method for managing three-dimensional printing | |
TWI709054B (en) | Building device and building method of prediction model and monitoring system for product quality | |
JP2022552981A (en) | Die-level product modeling without die-level input data | |
Bleicher et al. | Tooling systems with integrated sensors enabling data based process optimization | |
TWI729627B (en) | On-line size prediction method for fastener and on-line size prediction system for fastener | |
TWI684839B (en) | Method for diagnosing state of dies in fastener making machine and computer program product thereof | |
CN115329902A (en) | Method for analyzing and identifying abnormal data of product of numerical control plate rolling machine | |
D'Emilia et al. | Prediction of the remaining useful life of mechatronic systems, using internal sensors | |
CN101118423A (en) | Serviceability selecting method and system of virtual measuring prediction model | |
CN115187102A (en) | Model evaluation improvement method based on sinter quality prediction | |
CN112985318B (en) | Method and system for on-line prediction of fastener size | |
CN112388148A (en) | Simulated annealing algorithm-based welding parameter optimization method | |
Yang et al. | Intelligent Diagnosis of Forging Die based on Deep Learning |