Hasil Pencarian  ::  Simpan CSV :: Kembali

Abstrak :
Sistem tata udara presisi adalah sistem yang mengatur lingkungan udara yang cocok untuk peralatan ICT dalam kebinet ruang Datacenter yang khusus melayani penggunaan yang sangat penting dan kritis. Untuk mencegah kerusakan pada peralatan ICT dan pada media penyimpan akibat thermal shutdown, conductive anodic failures, hygroscopic dust failures, corrosion, dan short circuit, sistem tata udara presisi harus dapat mengendalikan temperatur dan kelembaban didalam kabinet, serta mampu beradaptasi terhadap perubahan temperatur akibat perubahan beban panas peralatan IT. Permasalahan yang dihadapi adalah bahwa sistem ini memiliki karakterisitik kompleks dan nonlinier yang sangat kuat yang sangat sukar dikendalikan oleh teknik kendali lanjut linier. Di dalam dissertasi ini diusulkan teknik kendali prediktif nonlinier baru yang disebut sebagai sistem kendali prediktif multi model berbasis supervisi untuk mengendalikan temperatur keluaran sistem tata udara presisi. Algoritma kendali tersusun dari tiga layer, yaitu layer optimasi kendali real-time untuk mengikuti perubahan sinyal acuan, layer adaptasi untuk menyesuaikan model PAC terhadap variasi beban panas, dan layer supervisi untuk menjamin kestabilan. Sistem PAC memiliki rancangan struktur baru yaitu penambahan kondenser sekunder yang berfungsi sebagai reheater untuk menurunkan RH keluaran evaporator. Prinsip kerja dan siklus kompresi uap sistem PAC diilustrasikan dalam psychrometric chart dan diagram enthalpi-tekanan. Model nonlinier sistem PAC diturunkan menggunakan teori pemodelan fisik berdasarkan prinsip konservasi energi dan kesetimbangan massa, dan kemudian dilinierisasi di sekitar titik kerja untuk mengembangkan model ruang keadaan orde-8 yang cocok untuk perancangan pengendali multivariabel. Kualitas model terlinierisasi dianalisa dari aspek respons transien, sifat controllability dan observability, dan interaksi antar variabel masukan-keluaran. Sebuah model nonlinier yang disebut sebagai multi model linier diusulkan dimana matriks parameter model diestimasi oleh algoritma identifikasi N4SID menggunakan himpunan data eksperimen masukankeluaran. Kontribusi utama dari dissertasi ini adalah multi model linier dapat diestimasi secara bertingkat dimana tiap tingkat identifikasi mempertahankan hubungan linier antar matriks parameter. Konsep model bertingkat ini juga mempermudah perancangan pengendali prediktif multi model dengan tetap mempertahankan optimasi kendali sebagai permasalahan quadratic programming. Mekanisme adaptasi pengendali prediktif dibentuk dengan memperbaharui model prediksi menggunakan algoritma N4SID rekursif. Untuk menjamin kestabilan sistem PAC dan menghindari fenomena bursting, algoritma deteksi ketidakcukupan eksitasi sinyal masukan dan monitoring sinyal diturunkan dalam persamaan rekursif, sehingga penambahan waktu komputasi tidak signifikan. Komputasi rekursif pada layer supervisi menjadi kontribusi terakhir. Kualitas model nonlinier hasil pemodelan fisik dan identifikasi bertingkat divalidasi melalui simulasi dan uji eksperimen baik secara kualitatif maupun kuantitatif. Sebagai indikator kinerja validasi model digunakan kriteria loss function dan kriteria final prediction error. Dari hasil uji simulasi dan eksperimen, hanya multi model linier menunjukkan kinerja model yang baik dari aspek kemampuan meniru karakteristik nonlinear sistem PAC dan nilai parameter analisa model yang baik, sehingga model ini cocok dipakai pada perancangan pengendali. Algoritma kendali yang diusulkan juga diverifikasi baik dalam kasus uji simulasi dan eksperimen, dan menunjukkan kemampuannya untuk menjejaki perubahan sinyal acuan.

---

Precision air conditioning (PAC) is a system that regulate air environment suitable for ICT equipments inside the cabinet of Datacenter room which serves very important and critical works. In order to overcome damage on ICT equipments and media storage due to thermal shutdown, conductive anodic failures, hygroscopic dust failures, corrosion, and short circuit, the PAC should be able to control the temperature and relative humidity inside the cabinet, and also able to adapt againts temperature change caused by interaction with humans, change of environment temperature, and change of heat load of ICT equipments. The problem encountered is that the PAC shows complex and highly nonlinear dynamics that is usually very difficult to control with linear advanced control systems. In this Dissertation, a new nonlinear predictive control called a supervision-based multi model predictive control to regulate the temperature outlet of PAC is presented. The proposed control algorithm consists of three layers, they are the optimization of real-time control layer for tracking the given set points, the adaptation layer for adjusting the PAC model againts variation of heat load, and the supervision layer for guarantee the closed loop stability. The work mechanism and vapourcompression cycle for the PAC system are illustrated using psychrometric chart and enthalpypressure diagram. A nonlinear model is derived using physical modeling theory based on the conservation of mass and energy balance principles, and then linearized about operating points for developing a 8th order state space model suited for multivariable control design. The quality of linearized model is analyzed in terms of response transient, controllability, observability, and interaction between input-output variables. A nonlinear model called multi linear model is proposed where the model parameter matrices are estimated by N4SID algorithm using a set of input-output data. The main contribution of this dissertation is that the multi linear model can be estimated using multi-stage subspace identification algorithm, where the relationship between model parameter matrices is still maintained linear. The concept of multi level models also simplify the design of multi model predictive controller retaining control optimization as a quadratic programming problem. The adaptation mechanism is performed by updating the prediction model using recursive N4SID algorithm. In order to guarantee system stability and to overcome bursting phenomena, a detection algorithm of less excitation signal and signals monitoring are derived in recursive forms, so that the control algorithm needs no significant additional computing power. The recursive computation in supervision layer is the last contribution for this dissertation. Quality of nonlinear model from physical modeling and system identification is validated through simulation and experimental test both qualitatively and quantitatively. Loss function and final prediction error are choosed as a performance criteria of model validation. From the simulation and experimental results, only the multi linear model shows good modeling performance in terms of ability to mimic the nonlinear behavior of PAC system and good parameter value of model analysis. The proposed control algorithm is also verified in case of simulation and experimental test showing its ability to track the set-point change.

Abstrak :
A multi model predictive control and proportional-integral controller switching (MMPCPIS) approach is proposed to control a nonlinear distillation column. The study was implemented on a multivariable nonlinear distillation column (Column A). The setpoint tracking and disturbance rejection performances of the proposed MMPCPIS were evaluated and compared to a proportional-integral (PI) controller and the hybrid controller (HC). MMPCPIS developed to overcome the HC?s limitation when dealing with very large disturbance changes (50%). MMPCPIS provided improvements by 27% and 31% of the ISE (integral of square error) for feed flow rate and feed composition disturbance changes, respectively, compared with the PI controller, and 24% and 54% of the ISE for feed flow rate and feed composition disturbance change, respectively, compared with HC.

Abstrak :
A multi model predictive control and proportional-integral controller switching (MMPCPIS) approach is proposed to control a nonlinear distillation column. The study was implemented on a multivariable nonlinear distillation column (Column A). The setpoint tracking and disturbance rejection performances of the proposed MMPCPIS were evaluated and compared to a proportional-integral (PI) controller and the hybrid controller (HC). MMPCPIS developed to overcome the HC’s limitation when dealing with very large disturbance changes (50%). MMPCPIS provided improvements by 27% and 31% of the ISE (integral of square error) for feed flow rate and feed composition disturbance changes, respectively, compared with the PI controller, and 24% and 54% of the ISE for feed flow rate and feed composition disturbance change, respectively, compared with HC.

Abstrak :
A Multiple Model Predictive Control (MMPC) approach is proposed to control a nonlinear distillation column. This control framework utilizes the best local linear models selected to construct the MMPC. The study was implemented on a multivariable nonlinear distillation column (Column A). The dynamic model of the Column A was simulated within MATLAB® programming and a SIMULINK® environment. The setpoint tracking and disturbance rejection performances of the proposed MMPC were evaluated and compared to a Proportional-Integral (PI) controller. Using three local models, the MMPC was proven more efficient in servo control of Column A compared to the PI controller tested. However, it was not able to cope with the disturbance rejection requirement. This limitation was overcome by introducing controller output configurations, as follows: Maximizing MMPC and PI Controller Output (called MMPCPIMAX). The controller output configurations of PI and single linear MPC (SMPC) have been proven to be able to improve control performance when the process was subjected to disturbance changes (F and zF). Compared to the PI controller, the first algorithm (MMPCPIMAX) provided better control performance when the disturbance sizes were moderate, but it was not able to handle a large disturbance of + 50% in zF.

Abstrak :
ABSTRAK
Pengendali swa tala dengan penempatan kutub merupakan Salah satu pengendali adaptif yang menggunakan estimasi dengan pendekatan linier untuk memperbaharui pengendalinya secara on-line. Pengendali swa tala ini mampu mempertahankan unjuk kerja lingkar tertutup sistem namun membutuhkan waktu yang cukup lama untuk melakukan adaptasi. Untuk mengatasi perubahan yang cepat, dapat diatasi dengan menggunakan beberapa model linier yang dapat merepresentasikan sistem untuk beberapa titik kerja melalui pendekatan linier dengan analisis secara off-line. Model yang akan dlaktifkan adalah model yang paling baik merepresentasikan sistem pada saat itu dengan menggunakan supervisor yang menentukan mekanisme 'switching. Pengendalian ini disebut pengendalian multi model.

Pengendalian multi model memiliki keterbatasan untuk penempatan model dalam database sehingga unjuk kerja lingkar tertutup akan menurun jika tidak ada satupun diantara model-model yang ada dalam database yang mampu mewakili kondisi kerja saat itu. Kelemahan - kelemahan dari metode pengendalian swa tala dan pengendalian multi model dapat diatasi dengan menggabungkan kedua metode tersebut sehingga menjadi pengendalian adaptif multi model yang mampu mempertahankan unjuk kerja lingkar tertutup sistem walaupun terjadi selang waktu perubahan set point yang cepat

---

Abstrak :
ABSTRACT
Traditional approaches in constructing response surface models typically ignore model uncertainty. If the relationship between the input factors and output characteristics of a process is very complex, traditional model building approaches may have limited effectiveness. In this paper, we propose a multi model ensemble and then implement this ensemble model to optimize the process performance. To form a multi model ensemble, we need to determine the weights of the different models, that is, values indicating relative importance among the models. To determine the weights, a hybrid weighting method is proposed, in which both global and local weighting methods are taken into account. Based on the hybrid weights of different models, a multi model ensemble is built and optimized. An example is illustrated to verify the effectiveness of the proposed approach. The results show that the proposed model can achieve more accurate predictive capability and that a better process improvement is reached.